(Chapman & Hall - CRC The R Series) Christian Ritz - Signe Marie Jensen - Daniel Gerhard - Jens Carl Streibig - Dose-Response Analysis Using R-CRC Press (2020)

Dose-Response
Analysis Using R
Dose-Response
Analysis Using R
Christian Ritz
Signe Marie Jensen
Daniel Gerhard
Jens Carl Streibig
CRC Press
Taylor & Francis Group
6000 Broken Sound Parkway NW, Suite 300
Boca Raton, FL 33487-2742

c 2020 by Taylor & Francis Group, LLC
CRC Press is an imprint of Taylor & Francis Group, an Informa business
No claim to original U.S. Government works
Printed on acid-free paper
International Standard Book Number-13: 978-1-138-03431-0 (Hardback)
This book contains information obtained from authentic and highly regarded sources. Rea-
sonable efforts have been made to publish reliable data and information, but the author
and publisher cannot assume responsibility for the validity of all materials or the conse-
quences of their use. The authors and publishers have attempted to trace the copyright
holders of all material reproduced in this publication and apologize to copyright holders if
permission to publish in this form has not been obtained. If any copyright material has not
been acknowledged please write and let us know so we may rectify in any future reprint.
Except as permitted under U.S. Copyright Law, no part of this book may be reprinted,
reproduced, transmitted, or utilized in any form by any electronic, mechanical, or other
means, now known or hereafter invented, including photocopying, microfilming, and record-
ing, or in any information storage or retrieval system, without written permission from the
publishers.
For permission to photocopy or use material electronically from this work, please access
www.copyright.com (http://www.copyright.com/) or contact the Copyright Clearance Cen-
ter, Inc. (CCC), 222 Rosewood Drive, Danvers, MA 01923, 978-750-8400. CCC is a not-
for-profit organization that provides licenses and registration for a variety of users. For
organizations that have been granted a photocopy license by the CCC, a separate system
of payment has been arranged.
Trademark Notice: Product or corporate names may be trademarks or registered trade-

marks, and are used only for identification and explanation without intent to infringe.
Library of Congress Cataloging-in-Publication Data
Names: Ritz, Christian, author.

Title: Dose-response analysis using R / by Christian Ritz, Signe M. Jensen,
Daniel Gerhard, Jens C. Streibig.
Description: Boca Raton, Florida : CRC Press, [2019] | Includes
bibliographical references and index.
Identifiers: LCCN 2019006744| ISBN 9781138034310 (hardback : alk.
paper) | ISBN 9781315270098 (e-book)
Subjects: LCSH: Drugs--Dose-response relationship. | Drugs--Testing--
Computer simulation.
Classification: LCC RM301.8 .R58 2019 | DDC 615.1--dc23
LC record available at https://lccn.loc.gov/2019006744
Visit the Taylor & Francis Web site at

http://www.taylorandfrancis.com
and the CRC Press Web site at

http://www.crcpress.com
Contents
Preface ix
1 Continuous data 1
1.1 Analysis of single dose-response curves . . . . . . . . . . . . 2
1.1.1 Inhibitory effect of secalonic acid . . . . . . . . . . . 2
1.1.1.1 Fitting the model . . . . . . . . . . . . . . . 3
1.1.1.2 Estimation of arbitrary ED values . . . . . 6
1.1.2 Data from a fish test in ecotoxicology . . . . . . . . . 6
1.1.3 Ferulic acid as an herbicide . . . . . . . . . . . . . . 9
1.1.4 Glyphosate in barley . . . . . . . . . . . . . . . . . . 13
1.1.5 Lower limits for dose-response data . . . . . . . . . . 19
1.1.6 A hormesis effect on lettuce growth . . . . . . . . . . 23
1.1.7 Nonlinear calibration . . . . . . . . . . . . . . . . . . 26
1.2 Analysis of multiple dose-response curves . . . . . . . . . . 31
1.2.1 Effect of an herbicide mixture on Galium aparine . . 31
1.2.2 Glyphosate and bentazone treatment of Sinapis alba 35
1.2.2.1 A joint dose-response model . . . . . . . . . 36
1.2.2.2 Fitting separate dose-response models . . . 39
2 Binary and binomial dose-response data 43

2.1.1 Acute inhalation toxicity test . . . . . . . . . . . . . 44
2.1.1.1 Link to ordinary logistic regression . . . . . 46
2.1.2 Tumor incidence . . . . . . . . . . . . . . . . . . . . . 47
2.1.3 Earthworm toxicity test: Abbott’s formula . . . . . . 49
2.1.4 Another earthworms toxicity test: Estimating the
upper limit . . . . . . . . . . . . . . . . . . . . . . . . 52
2.2.1 Toxicity of fluoranthene under different ultraviolet
radiation . . . . . . . . . . . . . . . . . . . . . . . . . 54
2.2.2 Toxicity of different types of selenium . . . . . . . . . 57
3 Count dose-response data 63

3.1.1 Counting number of fronds . . . . . . . . . . . . . . . 64
3.1.2 Counting offspring: Modeling hormesis . . . . . . . . 69
v
vi Contents
3.1.3 More counting offspring: Varying observation periods 73

3.2.1 Counting bacteria colonies: Wadley’s problem . . . . 79
4 Multinomial dose-response data 85

4.1 Trichotomous data . . . . . . . . . . . . . . . . . . . . . . . 86
4.1.1 Insecticide residues . . . . . . . . . . . . . . . . . . . 86
4.1.2 Effect of two arboviruses on chicken embryos . . . . . 90
5 Time-to-event-response data 95
5.1 Analysis of a single germination curve . . . . . . . . . . . . 97
5.1.1 Germination of Stellaria media seeds . . . . . . . . . 97
5.2 Analysis of data from multiple germination curves . . . . . 102
5.2.1 Time to death of daphnias . . . . . . . . . . . . . . . 104
5.2.1.1 Step 1 . . . . . . . . . . . . . . . . . . . . . 104
5.2.1.2 Step 2 . . . . . . . . . . . . . . . . . . . . . 107
5.2.2 A hierarchical three-way factorial design . . . . . . . 109
5.2.2.1 Step 1 . . . . . . . . . . . . . . . . . . . . . 112
5.2.2.2 Step 2 . . . . . . . . . . . . . . . . . . . . . 114
6 Benchmark dose estimation 119

6.1 Binomial dose-response data . . . . . . . . . . . . . . . . . 120
6.1.1 Pathogens in food . . . . . . . . . . . . . . . . . . . . 120
6.1.2 Chromosomal damage . . . . . . . . . . . . . . . . . 124
6.1.3 Tumor incidence continued: Integration of historical
data . . . . . . . . . . . . . . . . . . . . . . . . . . . 126
6.2 Continuous dose-response data . . . . . . . . . . . . . . . . 128
6.2.1 Toxicity of copper in an ecosystem with giant kelp . 129
6.2.2 Toxicity of an antituberculosis drug . . . . . . . . . . 133
6.3 Model averaging . . . . . . . . . . . . . . . . . . . . . . . . 135
6.3.1 Pathogens in food revisited . . . . . . . . . . . . . . 136
6.3.2 Toxicity of an antituberculosis drug revisited . . . . . 140
7 Hierarchical nonlinear models 145

7.1 Normally distributed dose-response data . . . . . . . . . . . 145
7.2 The R package medrc . . . . . . . . . . . . . . . . . . . . . 146
7.2.1 In vitro effects of the fungicide vinclozolin . . . . . . 147
7.2.2 Inhibition of photosynthesis in spinach . . . . . . . . 150
7.2.3 Herbicides with auxin effects . . . . . . . . . . . . . . 153
7.2.4 Drought stress resistance in Brassica oleracea . . . . 157
Appendix A Estimation 161

A.1 Nonlinear least squares . . . . . . . . . . . . . . . . . . . . . 162
A.2 Maximum likelihood estimation . . . . . . . . . . . . . . . . 163
A.2.1 Binomial dose-response data . . . . . . . . . . . . . . 164
A.2.2 Count dose-response data . . . . . . . . . . . . . . . 164
Contents vii
A.2.2.1 The Poisson distribution . . . . . . . . . . . 164

A.2.2.2 The negative-binomial distribution . . . . . 165
A.2.3 Time-to-event-response data . . . . . . . . . . . . . . 165
A.3 The transform-both-sides approach . . . . . . . . . . . . . . 166
A.4 Robust estimation . . . . . . . . . . . . . . . . . . . . . . . 166
A.5 Sandwich variance estimators . . . . . . . . . . . . . . . . . 167
A.6 Constrained estimation . . . . . . . . . . . . . . . . . . . . 168
A.7 Two-stage estimation for hierarchical models . . . . . . . . 168
A.7.1 Technical replicates . . . . . . . . . . . . . . . . . . . 169
A.7.2 Two-stage approaches . . . . . . . . . . . . . . . . . 169
A.7.3 Lindstrom-Bates algorithm . . . . . . . . . . . . . . . 170
A.8 Starting values and self-starter functions . . . . . . . . . . . 171
A.9 Confidence intervals . . . . . . . . . . . . . . . . . . . . . . 172
A.10 Prediction and inverse regression . . . . . . . . . . . . . . . 172
A.10.1 Effective dose . . . . . . . . . . . . . . . . . . . . . . 173
A.10.2 Relative potency . . . . . . . . . . . . . . . . . . . . 174
Appendix B Dose-response model functions 177

B.1 Log-logistic models . . . . . . . . . . . . . . . . . . . . . . . 178
B.1.1 Four-parameter log-logistic models . . . . . . . . . . 178
B.1.1.1 Three-parameter version . . . . . . . . . . . 179
B.1.1.2 Two-parameter version . . . . . . . . . . . . 179
B.1.1.3 E-max and Michaelis-Menten models . . . . 179
B.1.2 Extensions . . . . . . . . . . . . . . . . . . . . . . . . 180
B.1.2.1 Generalized log-logistic models . . . . . . . 180
B.1.2.2 A model with two slope parameters . . . . . 181
B.1.2.3 Hormesis models . . . . . . . . . . . . . . . 181
B.1.2.4 Two- and three-phase models . . . . . . . . 182
B.1.2.5 Fractional polynomial models . . . . . . . . 182
B.2 Log-normal models . . . . . . . . . . . . . . . . . . . . . . . 183
B.3 Weibull models . . . . . . . . . . . . . . . . . . . . . . . . . 184
B.3.1 Weibull type 1 models . . . . . . . . . . . . . . . . . 184
B.3.1.1 Exponential decay model . . . . . . . . . . 184
B.3.1.2 Other special cases . . . . . . . . . . . . . . 185
B.3.2 Weibull type 2 models . . . . . . . . . . . . . . . . . 185
B.3.2.1 Asymptotic regression . . . . . . . . . . . . 185
B.3.2.2 Other special cases . . . . . . . . . . . . . . 186
B.3.2.3 Generalized Weibull-2 model . . . . . . . . 186
B.4 Other types of models . . . . . . . . . . . . . . . . . . . . . 186
B.4.1 Gamma models . . . . . . . . . . . . . . . . . . . . . 186
B.4.2 Multistage models . . . . . . . . . . . . . . . . . . . . 186
B.4.3 NEC . . . . . . . . . . . . . . . . . . . . . . . . . . . 187
B.4.4 Biphasic models with a peak . . . . . . . . . . . . . . 187
B.5 Fixing parameters . . . . . . . . . . . . . . . . . . . . . . . 188
viii Contents
Appendix C R code for plots 191

C.1 Continuous dose-response data . . . . . . . . . . . . . . . . 191
C.1.1 Ferulic acid as an herbicide . . . . . . . . . . . . . . 192
C.2 Estimation of BMD . . . . . . . . . . . . . . . . . . . . . . 192
C.2.1 Pathogens in food . . . . . . . . . . . . . . . . . . . . 192
C.2.2 Toxicity of an antituberculosis drug . . . . . . . . . . 194
C.3 Hierarchical nonlinear models . . . . . . . . . . . . . . . . . 195
C.3.1 Inhibition of photosynthesis in spinach . . . . . . . . 196
C.3.2 Herbicides with auxin effects . . . . . . . . . . . . . . 196
C.3.3 Drought stress resistance in Brassica oleracea . . . . 197
Bibliography 199
Index 211
Preface
The history of dose-response analysis goes back many hundred years. One of
the more unusual applications is that numerous rulers had cupbearers who
tried the ruler’s food and drink to avoid poisoning and probably the demise
of the regent. The dose-response was the survival/health of the cupbearer.
In more recent times, dose-response analysis was applied to data from
controlled experiments where a limited number of doses of a toxic chemical
compound were to be compared to a control group (dose 0) in terms of binary
responses such as whether or not a treated insect was dead or alive after a
certain time period (Finney, 1949). Later dose-response analysis crystallized
into being a certain type of regression analysis. In the seminal work by Finney
(1971) it is explained how to carry out the estimation in the so-called probit re-
gression model through manual calculations. By the late 1970s dose-response
analysis had been extended to log-logistic models for continuous response
(Finney, 1979). In the beginning, such dose-response data were fitted through
linearization (e.g., Streibig, 1981, 1983). Later nonlinear estimation of such
models became available through add-ons and macros for spreadsheet pro-
grams (e.g., Vindimian et al., 1983; Caux and Moore, 1997). General-purpose
statistical software programs also included nonlinear estimation procedures
but without any specific focus on dose-response analysis.
By 2005 the first version of the extension package drc was developed for
the statistical programming environment R (R Core Team, 2018). Originally,
it was developed for nonlinear fitting of log-logistic models that were routinely
carried out in weed science (Ritz and Streibig, 2005). However, subsequently,
the package has been modified and extended substantially, mostly in response
to inquiries and questions from the user community. It has developed into a
veritable ecosystem for dose-response analysis (Ritz et al., 2015). Currently,
such extensive functionality for dose-response analysis does not exist in any
other statistical software. One of the problems that non-statistical scientists
were facing in the past was that guestimates of nonlinear regression parameters
had to be provided upfront before any estimation of parameters could take
place; this was an insuperable problem for many practitioners. To a very large
extent this problem has now been resolved in the package drc through the use
of so-called self-starter routines.
The development of dose-response analysis has undergone dramatic
changes from struggling with cumbersome more or less manual calculations
and transformations with pen and paper to the blink-of-an-eye estimation of
relevant parameters on any laptop.
ix
x Preface
A unified framework
The dose does not necessarily need to be a chemical compound. We define a
dose (metameter) as any pre-specified amount of biological, chemical, or ra-
diation stimuli or stress eliciting a certain, well-defined response. Other kinds
of exposure or stress could also be imagined, e.g., time elapsed in germination
experiments. However, in any case, the dose is a non-negative quantity.
Specifically, we define the response evoked by a specific dose as the quan-
tification of a biologically relevant effect, and as such, it is subject to random
variation. The most common type is a continuous response such as biomass,
enzyme activity, or optical density. A binary or aggregated binary (binomial)
response is also frequently used to describe results such as dead/alive, immo-
bile/mobile, or present/absent (Van der Vliet and Ritz, 2013). The response
may also be discrete as in a number of events observed in a specific time inter-
val such as a number of juveniles, offspring, or roots (Ritz and Van der Vliet,
2009). We will have more examples in later chapters.
A key feature of dose-response analysis is that the experimenter or re-
searcher has to have some a priori idea about the type of model function that
would be relevant for the analysis of her/his dose-response data. In principle,
many nonlinear model functions could be considered for describing how the
average response changes over the range of doses considered. In practice, only
a limited number of functions are used in the majority of applications. Specif-
ically, we will focus on modeling average trends through mostly s-shaped or
related biphasic functions. These functions reflect an a priori basic under-
standing of the causal relationship between the dose and the response, e.g.,
when a dose increases the response decreases between certain limits referred
to as the lower and upper limits, respectively. S-shaped functions have turned
out to be extremely versatile for describing various biological mechanisms;
one key feature is that model parameters provide useful interpretations of
observed effects within a biologically plausible framework. Specifically, dose-
response analysis is often used for screening and ranking of compounds using
estimated effective or lethal doses such as ED50 or LD50 (e.g., WHO, 2005).
The full specification of a statistical dose-response model involves both
specifying the parametric model function and assumptions about the distri-
bution of the responses, i.e., how they randomly fluctuate around the average
value determined by an assumed model function. Distributional assumptions
depend on the type of response observed. However, the same model functions
may be meaningful for different types of responses, and this is the unify-
ing feature of dose-response analysis: It involves dose-response models that
are a collection of statistical models that have a certain mean structure in
common. This is not a mathematical definition in any sense, but rather a
definition driven by applications, which actually makes sense for a statisti-
cal methodology. Consequently, dose-response models encompass a range of
statistical models that could be classified as nonlinear regression, generalized
Preface xi
(non)linear regression, and parametric survival analysis. Perhaps this is one

reason for dose-response analysis being neglected by many statisticians, as
already pointed out many years ago (Finney, 1979).
What this book is and is not about

Nowadays the term dose-response is used in many different contexts. This book
is about fitting and interpreting results obtained from fitting fully parametric
nonlinear dose-response (regression) models. In short, this is how we define
dose-response analysis and this book is meant to provide an introduction
to this kind of dose-response analysis. We do not cover some areas of dose-
response analysis that are still under development and where functionality is
not yet available in R, such as dose-response analysis of data from mixture
experiments and models for handling measurement error.
Some methods, which used to be part of a dose-response analysis, are
intentionally not covered in the book: goodness-of-fit and lack-of-fit tests and
other statistical tests that are used for assessing model assumptions. We prefer
using statistical tests for evaluating research questions about differences.
This book neither covers analysis of variance of dose-response data nor
estimation of linear trends, which are sometimes, and in some fields, referred
to as dose-response analysis. In the past, and occasionally still, s-shaped dose-
response trends or parts of them were approximated by means of linear regres-
sion models, e.g., parallel-line and slope-ratio assays, but such approaches are
hardly needed anymore as was already pointed out a long time ago (Vølund,
1978). We also do not cover semi- or non-parametric approaches for modeling
dose-response data. Finally, Bayesian methods for dose-response analysis are
still in their infancy and not covered either.
How to read the book

The first five chapters deal with different types of dose-response data, and
they may be read in arbitrary order as they are only loosely connected. How-
ever, there is a gradient such that more complex examples will be encountered
in the later chapters. The last chapters are more specialized, covering recent
advances, and therefore also more technical but still example-based. In exam-
ples, we will most often report estimates with corresponding standard errors
in brackets, unless otherwise stated. We provide an appendix with more tech-
nical details and explanations for the interested reader, but most examples
should be self-contained.
xii Preface
Above we praised the fact that modern dose-response analysis may be

carried out in much the same way as linear regression analysis. This is indeed
true, but as the estimation is intrinsically nonlinear, it can occasionally happen
that the analysis is also less smooth and the user will need to provide more
input. This is, in particular, the case when fitting more complex dose-response
models. We provide some examples in Chapter 7.
Sometimes we analyze examples in the chapters in this book in several
ways. This allows us to compare different approaches. In practice, a single
approach should ideally be decided upon by the scientist’s knowledge of the
underlying biological mechanisms. We hope this book will be helpful in ac-
quiring an informed opinion about when to use which approach.
Acknowledgment
We are fortunate in having some colleagues and experts, Florent Baty, Andrew
Kniss, Andrea Onofri, Janine Wong, and Ming Yi, who kindly agreed to read
sections or the entire manuscript. We are grateful for their valuable comments
and correction of the substance and language. We would stress, however, that
all these helpful people are in no way responsible for any mistakes which still
occur; these are ours alone.
1
Continuous data
In this chapter, we show examples of how to analyze dose-response data that

are continuous. It is a common type of dose-response data obtained in biol-
ogy, toxicology, and many other fields. Specifically, a continuous response is a
quantitative measurement that may take any value within a certain (possibly
unlimited) range and is usually expressed in a quantitative unit such as colour,
length, mass, and volume.
For continuous dose-response data, we will assume a statistical model
where each response value y may be decomposed into two additive contri-
butions. One contribution is determined or predicted by the corresponding
dose through the assumed model function f , i.e., it will be the same contri-
bution every time the same dose is applied; it is a fixed contribution for each
dose. This contribution corresponds to the average response for a given dose.
The other contribution is random, i.e., it will change from one replicate to
the next, modifying the average response value in an unpredictable way. The
randomness may, however, be characterized by a probability distribution that
quantifies how likely the contribution is to take values in any given interval
of response values. As these random contributions will modify the average
response by adding or subtracting some value, we say that the probability
distribution is centered around the mean (response). This statistical model
may be written in the following way:
yi = f (xi , β) + εi , i = 1, . . . , n (1.1)
with the fixed and random contributions adding up to the observed response
value for each pair of dose and response (xi , yi ), for a total of n measurements.
It is common to assume that the random contributions, the εi ’s in Equa-
tion (1.1), follow a mean-zero normal distribution with an unknown residual
standard deviation, which also is a model parameter to be estimated from
the data. The residual standard error is a measure of the variation between
measurements beyond what is explained by the assumed dose-response model
function. We will also address how to deal with dose-response data that do not
fully satisfy the above assumptions (see Subsection 1.1.3 and Subsection 1.1.4
for examples).
The model specification in Equation 1.1 relies on the assumption that the
variation between replicates is the same for all doses (referred to as variance
homogeneity). In this case, estimation may be carried out using nonlinear
1
2 Dose-response analysis using R
least squares (see Section A.1 for more details) as the dose-response model is
a special case of a nonlinear regression model (Ritz and Streibig, 2008).
In the examples below we will only specify the model function f and im-
plicitly assume that a statistical model is defined through Equation (1.1).
However, we will also address situations where the assumptions of normality
and variance homogeneity are not fulfilled.
In this chapter, we use the following extension packages:
library(drc)
library(devtools)
install_github("DoseResponse/drcData")
library(drcData)
library(boot)
library(lmtest)
library(metafor)
library(sandwich)
1.1 Analysis of single dose-response curves

In the first example there are no replicates per dose, i.e., only a single response
is obtained per dose. It can occur as a consequence of the experimental design,
which may be a good thing if it means that more doses and fewer replicates
were used. It can also happen because averages over the replicates were cal-
culated per dose; this is a waste of data and should be avoided: All replicates
should enter the analysis as shown in Subsection 1.1.2 and Subsection 1.1.3.
1.1.1 Inhibitory effect of secalonic acid

We want to carry out a dose-response analysis for the dataset secalonic
shown below.
secalonic
## dose rootl
## 1 0.000 5.5
## 2 0.010 5.7
Continuous data 3
## 3 0.019 5.4
## 4 0.038 4.6
## 5 0.075 3.3
## 6 0.150 0.7
## 7 0.300 0.4
Dose-response data were obtained from an experiment assessing the inhibitory

effect of secalonic acid on plant growth (Gong et al., 2004). The response values
(in the variable named rootl) are root lengths in cm, and the dose values
(dose) are in mM. Replicates were averaged per dose (as already pointed out
above, this is in general not a good idea).
1.1.1.1 Fitting the model

In weed science a four-parameter log-logistic model is often used to describe
dose-response relationships. We can use the model fitting function drm(),
which is the pivotal function in the package drc for fitting dose-response mod-
els. Specifically, the response is specified on the left-hand side of the tilde (∼)
as the root length (rootl). The independent variable, which is denoted dose,
is supplied to the right of the tilde. Next, we specify the dataset where the vari-
ables named dose and rootl are to be found using the argument data. The
argument fct = LL.4() specifies a four-parameter log-logistic model where
both lower and upper limits have to be estimated from the data. For contin-
uous dose-response data it is not necessary to specify the argument type as
the default is type = "continuous". We store the resulting model fit in an
object named secalonic.LL.4:
secalonic.LL.4 <- drm(rootl ~ dose,

data = secalonic,
fct = LL.4())
The first argument supplied to the function drm() is a model formula relating
the response to the predictor. The second argument data is specifying the
dataset where the variables rootl and dose are found. R will not automat-
ically look for variables in the dataset secalonic because they are not in
the search path and it is a good habit to specify the relevant dataset every
time a model is fitted. The third argument fct specifies the dose-response
model function that we want to fit. As there are 7 different doses, a four-
parameter model may easily be fitted (usually as many doses as parameters
are said to be required, but less will also do sometimes, possibly depending
on the choice of model function and the number of replicates). The built-in
function LL.4() in drc provides the four-parameter log-logistic model that
is commonly used in toxicology (see Section B.1 for more details). In short,
this model has four parameters: a lower limit, an upper limit, a parameter
corresponding to ED50, and a parameter for the relative slope at the dose
equal to ED50 (see Figure 1.1).
Note that upon execution of the above R lines, no output is produced.

All information on the model fitted to the data is stored in the object named
secalonic.LL.4, which is our choice for an informative name for the model
fit; you are invited to use other names.
A plot of the observed response (rootl versus dose) and the corresponding
fitted dose-response curve is obtained using the plot method.
Apart from the first argument (the model object secalonic.LL.4), the
arguments in plot govern the layout of the plot. While the arguments xlab,
ylab and ylim are self-explanatory (or see the help page ?par for graphical
parameters), the arguments bp and broken require a little more explanation.
broken indicates the presence or absence of a break in the dose axis, in or-
der to display that both 0 and the logarithm axis are shown and bp sets the
value on the dose axis where the axis breaks. A break is needed in situations
where zero concentrations/doses (x values) are combined with a logarithmic
x-axis. The default value of bp is usually quite sensible, but in this particular
plot(secalonic.LL.4,
bp = 1e-3, broken = TRUE,
ylim = c(0, 7),
xlab = "Dose (mM)",
ylab = "Root length (cm)")
7
6
Root length (cm)
5
4
3
2
1
0
0 0.01 0.1
Dose (mM)
FIGURE 1.1
The four-parameter log-logistic model fitted to dose-response data from the
dataset secalonic is plotted together with the original data (no replicates).
Continuous data 5
instance we must choose a smaller value to ensure that the bp value is smaller
than all positive concentrations/doses (otherwise some observations are not
displayed!). The argument log = "" may be used to switch off the default
logarithmic dose axis.
A summary of the fit is obtained using the summary method when applied
to the model fit secalonic.LL.4:
summary(secalonic.LL.4)
##
## Model fitted: Log-logistic (ED50 as parameter) (4 parms)
##
## Parameter estimates:
##
## Estimate Std. Error t-value p-value
## b:(Intercept) 2.6542086 0.6962333 3.8122 0.0317398 *
## c:(Intercept) 0.0917852 0.3747246 0.2449 0.8223012
## d:(Intercept) 5.5297495 0.2010300 27.5071 0.0001055 ***
## e:(Intercept) 0.0803547 0.0078829 10.1935 0.0020121 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
## Residual standard error:
##
## 0.2957497 (3 degrees of freedom)
The output shows the type of model that was fitted and the parameter esti-
mates for the four model parameters together with the corresponding (esti-
mated) standard errors. Briefly, the parameter b, c, d, and e refer to the slope
parameter for a dose equal to e, which is the dose resulting in a reduction
halfway between the upper limit d and the lower limit c, which is also called
ED50. We refer to Subsection B.1.1 for more details about the four-parameter
log-logistic model.
For each parameter, there are also t-values, which are parameter esti-
mates divided by their standard error, and the resulting p-values, looked up
in an appropriate t distribution; each of them corresponds to testing the null
hypothesis that the parameter is equal to 0 (not necessarily a relevant null
hypothesis to consider). The estimated residual variance is also shown, al-
though it is hardly reported in any publications, but it may still be useful
for understanding variation in the experiment. Possibly, the most interesting
parameter in the summary output is the parameter estimate for e; it is equal
to 0.08 (0.0079).
We also see that the estimated lower limit, which is equal to 0.0918 with a
standard error of 0.375, is not significantly different from zero (p-value = 0.82),
possibly indicating that a three-parameter log-logistic model (with an assumed
lower limit of 0) would also fit the data. However, such ad hoc data-driven
model reduction is not meaningful and should be avoided. Assuming a lower

limit equal to 0 should be based on a priori biological insights about the dose-
response experiment, i.e., will root length ever be exactly 0 even for very high
doses?
1.1.1.2 Estimation of arbitrary ED values

The function ED() calculates effective doses of your choice. Perhaps ED50 is
not always a relevant parameter. For instance, in ecotoxicology a safe level for
organisms may be defined using more protective levels such as ED10 or ED20.
See Section A.10.1 for more about effective doses.
Any ED value is a function of the model parameters (for the four-parameter
log-logistic model it is a function of the slope b and ED50), and it is possible
to estimate any ED value and also the corresponding standard error from the
model fit without having to re-parameterize and re-fit the model each time a
new ED value has to be estimated. For instance, ED10 and ED20 based on
the model fit secalonic.LL.4 are easily obtained using the following line:
ED(secalonic.LL.4, c(10, 20))
##
## Estimated effective doses
##
## Estimate Std. Error
## e:1:10 0.0351149 0.0078689
## e:1:20 0.0476628 0.0074229
The output shows estimated ED values and corresponding (approximate) stan-

dard errors. A 10% reduction in the root length relative to the estimated upper
and lower limits, which corresponds to a root length of 4.99, is achieved for a
dose of 0.035 (0.0079) mM. One way of understanding the calculation is by re-
versing it using the function predict to predict the fitted value corresponding
to the obtained dose:
predict(secalonic.LL.4,
data.frame(dose = ED(secalonic.LL.4, c(10), display = FALSE)))
## Prediction
## 4.985953
The argument display = FALSE switches off showing the output from ED().
1.1.2 Data from a fish test in ecotoxicology

The OECD (2006, p. 65) guideline provides test data from a 21-day fish test
using the test organism rainbow trout (Oncorhynchus mykiss). It is a standard
Continuous data 7
toxicity experiment with 7 concentrations: 0 and 6 non-zero concentrations.

The substance being tested at a range of concentrations is not known. There
are 10 replicates per concentration. However, for some of the higher doses,
several missing values occur so that the dataset contains only 61 observations
(instead of 70). We note that ignoring such missing values may lead to a bi-
ased results, depending on the mechanism causing missingness. The measured
response is the wet weight of the fish (in g). The dataset is available as the
data frame O.mykiss.
Following OECD (2006, p. 80) we fit the two-parameter exponential decay
model using the built-in function EXD.2(), which is a special case of the four-
parameter log-logistic model. We use summary to obtain a summary of the
model fit.
O.mykiss.EXD.2 <- drm(weight ~ conc,
data = O.mykiss,
fct = EXD.2(),
na.action = na.omit)
summary(O.mykiss.EXD.2)
##
## Model fitted: Exponential decay with lower limit at 0 (2 parms)
##
##
## d:(Intercept) 2.846794 0.092526 30.7674 < 2.2e-16 ***
## e:(Intercept) 111.738614 33.196876 3.3659 0.001347 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
Note that we specify the argument na.action to handle (in this case remove)
the 9 missing values mentioned earlier. The resulting model fit plotted together
with all data points (as indicated by the argument type = "all") is shown
in Figure 1.2. Alternatively, the default argument type = "average" would
imply plotting averages of replicates for each dose. Showing all data may be
helpful in evaluating the model fit, whereas showing averages may be a better
choice when reporting results.
The dataset is somewhat larger than in the previous example, so it makes
some sense to investigate if the model assumptions are fulfilled. Graphical
model checking can be based on the residual plot and QQ plot shown in Fig-
ure 1.3 and Figure 1.4, respectively. The visual assessment of the model fit
plot(O.mykiss.EXD.2,
broken = TRUE,
type = "all",
xlim = c(0, 500), ylim = c(0, 4),
xlab = "Concentration (mg/l)",
ylab = "Weight (g)")
3
Weight (g)
0 1 10 100
Concentration (mg/l)
FIGURE 1.2
A two-parameter exponential decay model fitted to dose-response data from
a fish test (the dataset named O.mykiss) with up to 10 replicates per concen-
tration (all replicates shown).
by the residual plot in Figure 1.3 indicates that the chosen model function
provides an appropriate description of the dose-response data because there is
random scatter above and below the reference line, which is the x axis, as the
residuals ought to be centered around 0. Furthermore, the residual plot shows
no indications of deviations from the model assumption of variance homogene-
ity as the scatter of points has the same spread across the entire range of the
x axis (which is similar to the range in the response values). Likewise, visual
assessment of the model fit by means of the QQ plot in Figure 1.4 indicates
that the normality assumption for the response values is appropriate because
the ranked (raw) residuals approximately match the corresponding expected
ordered values when assuming a normal distribution.
Continuous data 9
plot(fitted(O.mykiss.EXD.2),
residuals(O.mykiss.EXD.2))
abline(h = 0, lty = 2)
residuals(O.mykiss.EXD.2)
1.0
0.5
0.0
−1.0
2.0 2.2 2.4 2.6 2.8
fitted(O.mykiss.EXD.2)
FIGURE 1.3
Residual plot for the two-parameter exponential decay model fitted to the
dose-response data in the dataset O.mykiss. A reference line corresponding
to the x axis has been added.
As for the previous example, the natural next step would be to estimate
relevant ED values. However, we will not pursue this objective here.
1.1.3 Ferulic acid as an herbicide

As part of a larger experiment involving mixtures of several compounds, the
effect of the natural compound ferulic acid as a potential herbicide on growth
of perennial ryegrass was investigated (Inderjit et al., 2002). The resulting
dose-response data is found in the dataset ryegrass where the variable conc
is the concentration of ferulic acid in mM and rootl is the resulting root
length in cm. There are 7 concentrations, 6 replicates for concentration 0 and
3 replicates for all non-zero concentrations, in total 24 pairs of concentrations
and response values. We start fitting a four-parameter log-logistic model to
data as was also done in the original analysis reported by Inderjit et al. (2002).
qqnorm(residuals(O.mykiss.EXD.2))
abline(a = 0, b = sd(residuals(O.mykiss.EXD.2)))
1.0 Normal Q−Q Plot

Sample Quantiles
0.5
0.0
−1.0
−2 −1 0 1 2
Theoretical Quantiles
FIGURE 1.4
QQ plot for the two-parameter exponential decay model fitted to the dose-
response data in the dataset O.mykiss. A reference line corresponding to the
line with intercept 0 and slope equal to the empirical standard deviation of
the residuals has been added.
ryegrass.LL.4 <- drm(rootl ~ conc,

data = ryegrass,
fct = LL.4())
The fitted dose-response curve together with the data is shown in Figure 1.5.
We refer to Section C.1 for an example on how to make the plot by using the
package ggplot2 (Wickham, 2016).
Figure 1.5 shows that the variation in the response values is largest for
mid-range concentrations, whereas the variation is less for smaller concentra-
tions and much less for larger concentrations. Therefore, the model assump-
tion about variance homogeneity may be questionable. We can explore this
Continuous data 11
plot(ryegrass.LL.4,
broken = TRUE,
type = "all",
xlab = "Concentration (mM)",
ylab = "Root length (cm)")
8
Root length (cm)
0
0 1 10
Concentration (mM)
FIGURE 1.5
Four-parameter log-logistic model fitted to the dataset ryegrass (all repli-
cates shown).
observation further by means of graphical model checking: A residual plot

(Figure 1.6) also shows a tendency towards increasing variation in the mid-
range of response values (which corresponds to mid-range concentrations).
The residual plot shows that the scatter is evenly distributed above and below
the x axis, indicating that the chosen model function is adequately describ-
ing the average trend in the data. Therefore, we would not expect this slight
model misspecification (variance heterogeneity) to have much effect on the
estimated parameters, but it may affect the corresponding standard errors
to a larger extent. It is worth pointing out that these departures from the
model assumptions are unsurprising and may often be observed in a more or
less pronounced way as a consequence of biological mechanisms (less varia-
tion under severe stress) and the experimental design (dilution series used for
plot(fitted(ryegrass.LL.4), residuals(ryegrass.LL.4))
1.0
residuals(ryegrass.LL.4)
0.5
0.0
−1.0
2 4 6 8
fitted(ryegrass.LL.4)
FIGURE 1.6
Residual plot for the four-parameter log-logistic model fitted to the ryegrass
dataset.
the concentrations). The QQ plot (not shown) does not reveal any departures
from normality.
One way to adjust for the model misspecification is to use robust standard
errors, which do not require that a modified model is fitted (see Section A.5).
It is a remedy applicable to the already fitted model. We can use the function
coeftest() from the extension package lmtest and rely on functionality from
the package sandwich to obtain robust standard errors.
coeftest(ryegrass.LL.4,
vcov = sandwich)
##
## t test of coefficients:
##
## Estimate Std. Error t value Pr(>|t|)
## b:(Intercept) 2.98222 0.47438 6.2865 3.882e-06 ***
## c:(Intercept) 0.48141 0.12779 3.7672 0.001212 **
Continuous data 13
## d:(Intercept) 7.79296 0.15311 50.8976 < 2.2e-16 ***

## e:(Intercept) 3.05795 0.26741 11.4355 3.170e-10 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
To allow comparisons, the condensed summary output for the model fit (with
naive standard errors) is shown below.
coef(summary(ryegrass.LL.4))

## b:(Intercept) 2.9822191 0.4650623 6.412515 2.960347e-06
## c:(Intercept) 0.4814132 0.2121924 2.268758 3.450565e-02
## d:(Intercept) 7.7929583 0.1885672 41.327218 3.821922e-21
## e:(Intercept) 3.0579550 0.1857313 16.464401 4.267922e-13
The output shows a large decrease in the standard error for the parameter
estimate of c (the lower limit) and an increase in the standard error for the
estimated ED50. The standard errors for the estimates of b and d are less
affected. Given Figure 1.5, the changes in the standard errors make sense: In
the initial model fit where variance homogeneity was assumed, the increased
variation observed in the mid-concentration range (where ED50 is found) was
weighed down. The reduced variation observed for the larger concentrations
mostly determining the lower limit (the parameter c) was weighed up, reflect-
ing the assumption of variance homogeneity. The sandwich estimates do not
underlie the assumption of variance homogeneity and, therefore, captures the
actual variation in data to a higher degree. It is of course a balance. How
much variation do we want to accommodate? Ideally only variation caused by
some systematic effect or feature, not random variation.
The next example introduces yet another way to handle some types of
model misspecification.
1.1.4 Glyphosate in barley

We consider (unpublished) data from an experiment where different doses of
glyphosate (g a.i./ha) were applied to barley plants grown in a greenhouse. The
endpoint was biomass (g/pot). We again fit a four-parameter log-logistic model
to the data resulting in the fitted dose-response curve shown in Figure 1.7.
The variation in biomass seems to decrease with increasing doses.
head(barley)
## Dose weight
## 1 0.00000 57.2
## 2 0.00000 49.8
## 3 21.09375 62.2
## 4 21.09375 30.6
## 5 42.18750 40.9
## 6 42.18750 70.9
barley.LL.4 <- drm(weight ~ Dose,

data = barley,
fct = LL.4(),
na.action = na.omit)
summary(barley.LL.4)
##
##
##
## b:(Intercept) 9.7084 42.9166 0.2262 0.82430
## c:(Intercept) 11.1275 3.7803 2.9435 0.01068 *
## d:(Intercept) 52.0478 3.2487 16.0212 2.123e-10 ***
## e:(Intercept) 286.2600 209.6374 1.3655 0.19364
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
A residual plot also reveals the problem with the assumption of variance ho-
mogeneity: the variation in the response seems not to be constant (Figure 1.8).
In some cases, a transformation could help to achieve variance homogeneity.
The logarithm transformation would be one option as it is often success-
ful in removing the pattern observed in Figure 1.7: small variation in small
predicted values and large variation in large predicted values. However, in
order to preserve the assumed dose-response model, both the response and
the predicted values based on the model have to be transformed. This tech-
nique was used for the logarithm transformation by Streibig (1983) and, in
general terms, it is described as the transform-both-sides approach (Carroll
and Ruppert, 1988, Chapter 4). It is also possible to search for the optimal
transformation (within a family of transformations). We will consider Box-Cox
transformations, which are power functions with negative and positive expo-
nents (but also including the logarithm); see Section A.3 for more details. It
should be mentioned that there is some evidence that the transform-both-
sides approach may lead to inappropriate fitted dose-response curves in some
Continuous data 15
plot(barley.LL.4,
broken = TRUE,
type = "all",
xlab = "Dose (g a.i./ha)",
ylab = "Biomass (g/pot)")
70
60
Biomass (g/pot)
50
40
30
20
10
0 10 100 1000
Dose (g a.i./ha)
FIGURE 1.7
The four-parameter log-logistic model fitted to the barley data (barley).
cases (Ritz and Van der Vliet, 2009): so it is not an approach that we would
like to recommend for general use.
We can obtain the model fit for the optimal so-called Box-Cox transfor-
mation by refitting the model using the function boxcox(). The argument
method specifies that we want to estimate and apply the optimal power ex-
ponent, which is determined from the slightly more general one-way ANOVA
model rather than directly from the dose-response model (as there is a com-
putational gain in doing so).
barley.LL.4.bc <- boxcox(barley.LL.4,

method = "anova",
plotit = FALSE)
summary(barley.LL.4.bc)
plot(fitted(barley.LL.4), residuals(barley.LL.4))
20
residuals(barley.LL.4)
10
0
−10
−20
10 20 30 40 50
fitted(barley.LL.4)
FIGURE 1.8
Residual plot for the four-parameter log-logistic model fitted to the barley
data (barley).
##
##
##
## b:(Intercept) 4.6567 4.3207 1.0778 0.299362
## c:(Intercept) 10.5303 1.1542 9.1231 2.875e-07 ***
## d:(Intercept) 51.4583 5.4983 9.3590 2.109e-07 ***
## e:(Intercept) 250.8884 52.3177 4.7955 0.000285 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
Continuous data 17

##
## Non-normality/heterogeneity adjustment through Box-Cox
## transformation
##
## Estimated lambda: -0.25
## Confidence interval for lambda: [-0.576, 0.370]
The optimal power exponent (λ) for the Box-Cox transformation is –0.25,
which is significantly different from 1 (no transformation) as the corresponding
confidence interval does not include 1. However, the value 0 (corresponding to
the log-transformation) is included in the confidence interval. Except for the
slope parameter b, none of the parameter estimates changed dramatically due
to the transformation.
Alternatively, we may also directly specify the appropriate power exponent
using the argument bcVal in the model specification. This approach of making
an informed choice is to be preferred over estimating the optimal transforma-
tion as we did above. Perhaps only a few transformations are really relevant in
practice, e.g., the logarithm transformation and, occasionally, the square root
transformation. The drawback of this approach might be that some a priori
information about the distribution of the response is required (possibly from
other similar experiments).
barley.LL.4.log <- drm(weight ~ Dose,
data = barley,
fct = LL.4(),
na.action = na.omit,
bcVal = 0)
summary(barley.LL.4.log)
##
##
##
## b:(Intercept) 6.6523 8.8818 0.7490 0.466271
## c:(Intercept) 10.7738 1.1385 9.4628 1.843e-07 ***
## d:(Intercept) 51.1127 4.4105 11.5890 1.461e-08 ***
## e:(Intercept) 269.2678 76.8634 3.5032 0.003513 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
##
## Non-normality/heterogeneity adjustment through Box-Cox
## transformation
##
## Specified lambda: 0
We can add the fitted dose-response curve obtained using the transform-both-
sides approach with the logarithm transformation in the plot of the fitted
dose-response curve using untransformed data. The two fitted curves agree
quite well (see Figure 1.9).
plot(barley.LL.4.log,
add = TRUE,
lty = 2)
Finally, we may choose to stay with our initial model but using sandwich
estimates for the standard errors as in the example above. It should, however,
be noticed that the sandwich standard errors rely on the mean structure, i.e.,
70
60
Biomass (g/pot)
50
40
30
20
10
0 10 100 1000
Dose (g a.i./ha)
FIGURE 1.9
The four-parameter log-logistic model fitted to the barley data (barley), both
untransformed data (solid line) and when using a transform-both-sides ap-
proach with the logarithm transformation (dashed line).
Continuous data 19
the model function, being correctly specified, which may be hard to judge for
the present data example with sparse information in the middle of the curve.
coeftest(barley.LL.4,
vcov. = sandwich)
##
##
## b:(Intercept) 9.7084 33.8435 0.2869 0.7784
## c:(Intercept) 11.1275 1.1026 10.0918 8.341e-08 ***
## d:(Intercept) 52.0478 4.0530 12.8417 3.899e-09 ***
## e:(Intercept) 286.2600 165.3541 1.7312 0.1054
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
In summary both the transform-both-sides approach and the use of robust

standard errors lead to sensible changes in the standard errors of the parameter
estimates for the lower and upper limits. The standard error became smaller
for the lower limit and it became larger for the upper limit as compared to the
standard errors from the initial model barley.LL.4, which assumed variance
homogeneity. For the slope and ED50 parameter estimates, all approaches lead
to smaller standard errors, but they are still mostly large compared to the
magnitude of the corresponding estimate, indicating that there is essentially
no information in the data to determine the slope and EC50 parameters. At
best, only EDxx for large percentages xx may be reliably estimated and such
quantities are usually not that relevant.
1.1.5 Lower limits for dose-response data

For growth-related measurements, the response sometimes does not approach
a lower limit of 0 even though larger and larger doses are applied. One expla-
nation is that an organism being exposed to a toxic compound at a certain
time during its development has already had a growth before exposure and
this growth may or may not be preserved for a longer time depending on the
mode of action of the toxic compound. Another explanation could be drug
resistance. Consequently, the lower limit of the corresponding dose-response
curve may be positive, reflecting growth before exposure. The two following
examples will illustrate the concept. However, there may also be other reasons
for the absence or presence of a lower limit such as a systematic measurement
error that prevents observation of response values asymptotically approaching
0. In any case, such a priori information about one aspect of the dose-response
relationship should ideally be incorporated when choosing the model function.
Below we consider two illustrative examples.
The first example uses data from an experiment where ryegrass plants
were treated with a mixture of commercial herbicides that prevents further
growth but leaves the plants green and viable for a long time. The herbi-
cides are acetolactate synthase inhibitors that do not block photosynthesis
per se, but prevent the photosynthetic production from making the plant
grow. The biomass of the plants was measured on the day of spraying (where
the mixture would have no visible effect) and 15 days after spraying (for a
range of doses). The data, which are unpublished, are found in the dataset
ryegrass2.
head(ryegrass2)
## dose biomass day

## 1 0 77.5 0
## 2 0 78.0 0
## 3 0 75.0 0
## 4 0 214.4 15
## 5 0 215.4 15
## 6 0 227.5 15
We fit a four-parameter log-logistic model to the subset of data containing

information on the measurements 15 days after spraying.
ryegrass2.LL.4 <- drm(biomass ~ dose,

data = subset(ryegrass2, day == "15"),
fct = LL.4())
Figure 1.10 shows the fitted dose-response curve with a horizontal line added
showing the mean biomass level before spraying (at day 0).
In this instance, the biomass production (growth) at the time of spraying
(day 0) closely corresponds to the lower limit due to the mode of action of the
herbicide.
The second example involves the contact herbicide diquat, which was ap-
plied to red fescue, and after 16 days the biomass was measured (unpublished
data). The herbicide is a desiccant that kills the plant instantaneously. It
means the degradation of the plant commences immediately after the spray-
ing. Therefore, for high doses, we would expect biomass to approach 0.
head(red.fescue, 4)
## dose day biomass

## 1 0 0 45.0
## 2 0 0 69.0
## 3 0 16 137.0
## 4 0 16 102.0
Continuous data 21
plot(ryegrass2.LL.4,
broken = TRUE,
ylim = c(0, 250),
ylab = "Weed biomass (g/pot)")
abline(h = mean(subset(ryegrass2, day == "0")[["biomass"]]),

lty = 2)
250
Weed biomass (g/pot)
200
150
100
50
0 1 10 100
Dose (g a.i./ha)
FIGURE 1.10
Fitted four-parameter log-logistic model for biomass of ryegrass 15 days after
being sprayed with a mixture of herbicides that stop growth but leave af-
fected plants green (using the dataset ryegrass2). The dashed line indicates
biomass on day 0, the day when the experiment was sprayed with the herbicide
mixture.
We fit a four-parameter log-logistic model.
red.fescue.LL.4 <- drm(biomass ~ dose,

data = subset(red.fescue, day == "16"),
fct = LL.4())
plot(red.fescue.LL.4,
broken = TRUE,
ylim = c(0, 150),
ylab = "Biomass (g/pot)")
abline(h = mean(subset(red.fescue, day =="0")[["biomass"]]),

lty = 2)
abline(h = coef(red.fescue.LL.4)["c:(Intercept)"], lty = 3)
150
Biomass (g/pot)
100
50
0 100 1000
Dose (g a.i./ha)
FIGURE 1.11
Fitted four-parameter log-logistic model for biomass of red fescue 16 days after
being sprayed with the desiccant herbicide diquat that kills the plant quickly.
Dashed broken line indicates the biomass level at the time of spraying (day
0) and the dotted line shows the estimated lower limit after 16 days.
Figure 1.11 shows the fitted dose-response curve together with a horizontal
line representing the level of biomass at the time of spraying (day 0), which
does not coincide with the estimated lower limit that the mean response
is approaching for large doses. In this case, it reflects a different mode of
action of the herbicide. It is worth noting that the data still seem to indi-
cate that there is a positive lower limit, but it is not reflecting growth before
exposure.
Continuous data 23
1.1.6 A hormesis effect on lettuce growth

Hormesis is a stimulatory response to low levels of stress, such as low concen-
trations of organic or inorganic chemicals or radiation. Hormesis has been ob-
served for various organisms such as animals, humans, microbiota, and plants;
see Cedergreen et al. (2005) and Beckon et al. (2008) for more details.
We re-analyze data from an experiment where lettuce plants were grown in
nutrient solutions containing different concentrations of dissolved isobutylal-
cohol (mg/L) (van Ewijk and Hoekstra, 1993). Shoot weight (g) was measured
after 21 days. Seven concentrations (including the control) were used. For each
concentration duplicate measurements were obtained. The data are found in
the dataset lettuce. The entire dataset is shown below.
lettuce
## conc weight
## 1 0.00 1.126
## 2 0.00 0.833
## 3 0.32 1.096
## 4 0.32 1.106
## 5 1.00 1.163
## 6 1.00 1.336
## 7 3.20 0.985
## 8 3.20 0.754
## 9 10.00 0.716
## 10 10.00 0.683
## 11 32.00 0.560
## 12 32.00 0.488
## 13 100.00 0.375
## 14 100.00 0.344
The variables conc and weight contain the concentration and corresponding
weight measurements. A scatterplot of the data is shown in Figure 1.12. A
warning is issued because of the argument log = "x", which imposes a log-
arithmic concentration axis that allows us to better appreciate the hormesis
effect: there is a fairly clear inverse j-shape with an increase at low concentra-
tions before showing the expected decreasing trend for higher concentrations.
We fit the four-parameter Brain-Cousens hormesis model with the lower
limit fixed at 0 as we assume that there will not be any growth for high
concentrations. The special feature of this hormesis model, as well as other
hormesis models, is the additional model parameter that quantifies the degree
of hormesis. For the Brain-Cousens model, this parameter is denoted f , and
the interpretation is that f = 0 implies no hormesis effect, whereas f > 0
implies some degree of hormesis, the larger the value the larger the effect.
However, the actual value of f cannot directly be understood by the scale of
the response. The parameters b and d have the same interpretation as for the
plot(weight ~ conc,
data = lettuce, log = "x",
xlab = "Isobutylalcohol concentration (mg/L)",
## Warning in xy.coords(x, y, xlabel, ylabel, log): 2 x values

<= 0 omitted from logarithmic plot
1.2
1.0
Weight (g)
0.8
0.6
0.4
0.5 2.0 5.0 20.0 100.0
Isobutylalcohol concentration (mg/L)
FIGURE 1.12
Data in the lettuce dataset showing a hormesis effect.
log-logistic models: b is proportional to the slope at the concentration equal

to e, which, however, is not equal to EC50, and d is the upper asymptote or
limit of the dose-response curve (see Equation B.12 in Subsection B.1.2 for
more details). This model was also used in the original analysis by van Ewijk
and Hoekstra (1993). The built-in model function BC.4() invokes this model.
lettuce.BC.4 <- drm(weight~conc,

data = lettuce,
fct = BC.4())
Figure 1.13 shows the original data with the fitted dose-response curve super-
imposed. The model appears to fit the data well.
The summary output shows that the hormesis parameter f is not signifi-
cantly different from 0 (p-value is 0.129), which means that we cannot reject
the hypothesis of no hormetic effect using the dataset lettuce.
Continuous data 25
plot(lettuce.BC.4,
broken = TRUE,
xlab = "Isobutylalcohol concentration (mg/L)",
1.2
1.0
Weight (g)
0.8
0.6
0.4
0 1 10 100
Isobutylalcohol concentration (mg/L)
FIGURE 1.13
The fitted concentration-response curve for the four-parameter Brain-Cousens
model fitted to the dataset lettuce, shown together with the data (average
per concentration).
summary(lettuce.BC.4)
##
## Model fitted: Brain-Cousens (hormesis) with lower limit fixed
## at 0 (4 parms)
##
##
## b:(Intercept) 1.282812 0.049346 25.9964 1.632e-10 ***
## d:(Intercept) 0.967302 0.077123 12.5423 1.926e-07 ***
## e:(Intercept) 0.847633 0.436093 1.9437 0.08059 .
## f:(Intercept) 1.620703 0.979711 1.6543 0.12908
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
More importantly, the presence of a hormesis effect will impact the estimation
of any EC value. Ignoring the hormesis effect will typically lead to too small
(conservative) estimated EC values (Cedergreen et al., 2005). Therefore it may
be relevant to fit a hormesis model even though there is per se no interest in
assessing or describing the hormesis effect.
1.1.7 Nonlinear calibration

Sometimes a dose-response experiment is carried out with the aim to use the
resulting model fit to predict doses from response values observed in sub-
sequent experiments where doses are unknown. The former experiment is a
calibration experiment, whereas the latter ones are prediction experiments.
This kind of dose-response analysis is referred to as nonlinear calibration, and
it involves inverse regression. It is basically the same as comparing a response
of an unknown sample to a standard curve.
We revisit the following data example first considered by Racine-Poon
(1988): To explore degradation of an agrochemical in soil samples it was ap-
plied in 0-10 cm depth in 7 different doses, including dose 0 (g/ha). The
observed response is the weight (mg) of a nasturtium plant after 3 weeks.
This calibration experiment was repeated 6 times (so an improved analysis
would model variation between the 6 runs explicitly).
The first 12 lines of the dataset look like this:
head(nasturtium, 12)
## conc rep wt
## 1 0.000 1 920
## 2 0.025 1 919
## 3 0.075 1 870
## 4 0.250 1 880
## 5 0.750 1 693
## 6 2.000 1 429
## 7 4.000 1 200
## 8 0.000 2 889
## 9 0.025 2 878
## 10 0.075 2 825
## 11 0.250 2 834
## 12 0.750 2 690
Continuous data 27
As in the original analysis, we fit a three-parameter log-logistic model:
nasturtium.LL.3 <- drm(wt ~ conc, data = nasturtium, fct = LL.3())
summary(nasturtium.LL.3)
##
## Model fitted: Log-logistic (ED50 as parameter) with lower
## limit at 0 (3 parms)
##
##
## b:(Intercept) 1.350256 0.113557 11.891 1.518e-14 ***
## d:(Intercept) 897.862776 13.844884 64.852 < 2.2e-16 ***
## e:(Intercept) 1.576200 0.095694 16.471 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
Backfitting is a quick way to see how good the calibration model is. We use
the function backfit() in drc that takes the model fit object as the only
argument.
backfit(nasturtium.LL.3)
## Warning in log(exp(-tempVal/parmVec[5]) - 1): NaNs produced

## dose Estimate
## [1,] 0.000 NaN
## [2,] 0.025 0.1152911
## [3,] 0.075 0.1522799
## [4,] 0.250 0.2418827
## [5,] 0.750 0.7209820
## [6,] 2.000 2.0729493
## [7,] 4.000 3.8933864
The agreement is good, although for low doses (large response values) esti-
mated doses are too large. The warning message may be ignored as it simply
means it is not possible to estimate dose 0.
Let us assume that a prediction experiment resulted in response values 690,
693, and 722; this is similar to but not identical to what Racine-Poon (1988)
did. For each single response value, we can use the function ED() for estimating
the corresponding dose with a 95% confidence interval. For instance, for the
value 690 we get:
ED(nasturtium.LL.3, 690, type = "absolute", interval = "delta")
##
##
## Estimate Std. Error Lower Upper
## e:1:690 0.648194 0.070708 0.505174 0.791214
Note that the argument type = "absolute" is essential. By default, the argu-
ment is type = "relative", which is suitable for estimating effective doses.
The 95% confidence interval of the estimated dose is [0.505, 0.791]. However,
this interval is likely somewhat too narrow as it does not incorporate the vari-
ation in the response value. Likewise, we could also estimate the dose based
on the mean of the response values:
ED(nasturtium.LL.3, (690+693+722)/3, type = "absolute",

interval = "delta")
##
##
## e:1:701.666666666667 0.613385 0.069337 0.473137 0.753633
One way to incorporate variation in the response values, but also utilize that
three values were recorded, is through a bootstrap approach where estimated
doses are repeatedly computed for randomly sampled values from a normal
distribution assumed for the mean of the three response values. Specifically,
we will assume that the mean of 690, 693, and 722 follow a normal
√ distribution
with mean (690 + 693 + 722)/3 and standard deviation 56/ 3 where 56 is the
residual standard error from the above model fit.
To do a bootstrap, we use the extension package boot. As a first attempt
we will do a crude parametric bootstrap, which will incorporate the variation
in the response values, but it will not propagate the uncertainty on parameter
estimates when doing inverse regression. The key function boot() is specified
as follows: The first argument is the data, i.e., the mean response. The second
argument is the R function that converts the data value into an estimated
concentration; this is essentially ED() but being wrapped up in a function en-
vironment to allow only one value to vary. The third argument is the function
that returns randomly sampled values from the assumed normal distribution
of the mean; it is relying on the R function rnorm(). The fourth argument is
Continuous data 29
needed when giving the third argument. The last argument specifies the num-
ber of times to randomly draw mean values (it is common to use 1000). The
function set.seed() is used to ensure that the bootstrap procedure results
in the same results every time (based on the same pseudo-random numbers).
set.seed(201806061)
nasturtium.boot.res1 <- boot((693+722+690)/3,
statistic = function(simyVal){
ED(nasturtium.LL.3,
simyVal,
type = "absolute",
display = FALSE)[1]
},
ran.gen = function(yVal, mle){
rnorm(1, yVal, 55.55935/sqrt(3))
},
sim = "parametric",
R = 1000)
summary(nasturtium.boot.res1)
## R original bootBias bootSE bootMed

## 1 1000 0.61338 0.0028725 0.098303 0.61955
A 95% percentile bootstrap confidence interval may be obtained using the
function boot.ci(), also from the package boot. Specifically, we will use the
basic bootstrap confidence interval, which includes the middle 95% of the
estimated dose values based on generated response values, i.e., the middle 950
values among the 1000 being generated by the function boot() (Davison and
Hinkley, 1997).
boot.ci(nasturtium.boot.res1, type = "basic")
## BOOTSTRAP CONFIDENCE INTERVAL CALCULATIONS

## Based on 1000 bootstrap replicates
##
## CALL :
## boot.ci(boot.out = nasturtium.boot.res1, type = "basic")
##
## Intervals :
## Level Basic
## 95% ( 0.4200, 0.8032 )
## Calculations and Intervals on Original Scale
The resulting 95% confidence interval becomes [0.42, 0.8], which is somewhat
wider than the one we found by simply using ED(), [0.47, 0.75].
An improved parametric bootstrap approach would not only generate val-

ues for the mean response in the prediction model but also for all response
values used in the calibration model, as such an approach would ensure that
a 95% confidence interval of the dose of interest is based on all the variation
present in the analysis. For each generated set of 42 + 3 = 45 response values,
the three-parameter log-logistic model should be fitted and the estimated con-
centration obtained from inverse regression; this is what is happening below
in the second argument, based on randomly generated response values.
set.seed(201806062)
nasturtium.boot.res2 <- boot(c(nasturtium[["wt"]], 690, 693, 722),
statistic = function(yValues){
ED(drm(head(yValues, -3) ~ conc,
data = nasturtium,
fct = LL.3()),
mean(tail(yValues, 3)),
type = "absolute",
display = FALSE)[1]
},
ran.gen = function(yVal, mle){
rnorm(42+3,
c(fitted(nasturtium.LL.3),
690, 693, 722), 55.55935)
},
sim = "parametric",
R = 1000)
boot.ci(nasturtium.boot.res2, type = "basic")
## BOOTSTRAP CONFIDENCE INTERVAL CALCULATIONS

## Based on 1000 bootstrap replicates
##
## CALL :
## boot.ci(boot.out = nasturtium.boot.res2, type = "basic")
##
## Intervals :
## Level Basic
## 95% ( 0.3780, 0.8184 )
## Calculations and Intervals on Original Scale
The resulting 95% bootstrap confidence interval is slightly wider than the
previous one, implying that not much additional variation is picked up by
repeating all the steps of the nonlinear calibration. The above results are sim-
ilar to the ones derived by Racine-Poon (1988) through a completely different
approach.
Continuous data 31
1.2 Analysis of multiple dose-response curves

In this section, we consider two data examples with two treatment groups.
In both examples below, the main interest lies in comparing treatments in
terms of some feature of the dose-response curves that may be captured by
model parameters or a derived parameter. It is common to compare ED50
values between treatments, either in terms of differences or ratios depending
on which quantification is most relevant, e.g., in toxicology and weed science
it is often the ratio that is of interest.
In the first example, there is only a single dose-response sub-experiment
for each of two treatments, implying that there is no way to separate random
variation between dose-response curves (due to various unobserved differences
in the execution of each dose-response sub-experiment) from systematic dif-
ferences between treatments. Consequently, only weak claims on treatment
differences may be made based on such an experimental design. In the second
example, there are two replicated dose-response sub-experiments for each of
two treatments, allowing a stronger claim about a possible treatment differ-
ence. In a later chapter (Chapter 7) we will show how to analyze dose-response
data for more complex, possibly multiway factorial, experimental designs.
1.2.1 Effect of an herbicide mixture on Galium aparine

Small plants of cleavers (Galium aparine), growing in pots (one plant per pot)
in a large greenhouse, were sprayed with the herbicide phenmedipham either
alone or in a mixture with methyl oleate. For both herbicide treatments the
same 11 doses (g/ha) were used. There were 10 replicates per dose and per
treatment except for dose 0 where there were 40 replicates. Dose 0 was shared
between the two treatments, but in the dataset (G.aparine) the controls were
assigned to one of the treatments (it does not matter which one as long as the
dose-response model assumes a shared control). The plants were allowed to
grow in the greenhouse for 14 days after treatment, and then the dry matter
(weight in mg/pot) was measured. The experiment is described in more detail
by Cabanne et al. (1999). The first 6 lines of the dataset are shown below.
head(G.aparine)
## dose drymatter treatment

## 1 0 1146 1
## 2 0 1005 1
## 3 0 756 1
## 4 0 1108 1
## 5 0 956 1
## 6 0 989 1
The variable drymatter contains the measured dry weight per pot, the vari-
able dose contains doses of the two herbicide treatments, and the variable
treatment encodes the two herbicide treatments: 1 corresponds to phen-
medipham alone and 2 corresponds to the mixture of phenmedipham and
methyl oleate; the latter is an adjuvant that does not have any effect in the
plants.
For each of the two herbicide treatments, a single dose-response sub-
experiment with replicates per dose was carried out. Such an experimental
design implies that the herbicide treatment effects and the sub-experiment
effects (if present) are confounded; had there been multiple dose-response
sub-experiments per herbicide treatment these two effects could have been
disentangled.
We fit a joint model based on the entire dataset including both herbicide
treatments. Actually, as there is a shared control group, there is no other op-
tion than a joint model to avoid that the control group counts twice in the
analysis. Again, we use the model fitting function drm(), including the addi-
tional argument named curveid that provides information on the grouping
in the data and also including the argument pmodels that for each of the
four model parameters b, c, d, and e (in that order) specifies if the grouping
should be applied (∼treatment) or no grouping should be used, i.e., a single
common model parameter shared by all groups (∼1). As there is a shared
control group we have to insist on a shared parameter d for the upper limit.
The model specification looks like this.
G.aparine.LL.4 <- drm(drymatter ~ dose,

curveid = treatment,
data = G.aparine,
fct = LL.4(),
pmodels = list(~treatment,
~treatment,
~1,
~treatment))
So for each model parameter, we use a model formula specification similar

to what is used for lm() in R. These model formulas are combined in a list;
alternatively a data frame could be supplied. This means that we can compare
the two herbicide treatments in terms of any of the model parameters b, c, or
e. However, in practice, only comparisons of ED50 values (the parameters e
in this example) are of interest.
Figure 1.14 shows the fitted dose-response curves together with average
data per dose. The model seems to describe the dose-response relationships
quite well. A residual plot and QQ plot could be used for a more detailed
evaluation of the model assumptions in the same way as seen for a single
dose-response curve in Subsection 1.1.2.
Continuous data 33
plot(G.aparine.LL.4,
broken = TRUE,
xlab = "Dose (g/ha)",
ylab = "Dry weight (mg/pot)")
1400
1200 1
Dry weight (mg/pot)
2
1000
800
600
400
200
0 1 10 100 1000
Dose (g/ha)
FIGURE 1.14
The fitted dose-response curves for the four-parameter log-logistic model (with
a shared control group) fitted to the dataset G.aparine. Average data per
concentration are also shown.
As we assume a four-parameter log-logistic model per herbicide treatment,

but with the constraint that the upper limit parameter is shared, we estimate
a total of 7 model parameters (shown in the summary output below).
summary(G.aparine.LL.4)
##
##
##
## b:(Intercept) 1.61291 0.33330 4.8392 2.373e-06 ***
## b:treatment2 0.13809 0.37301 0.3702 0.7115667
## c:(Intercept) 509.50242 23.25903 21.9056 < 2.2e-16 ***

## c:treatment2 -357.58391 34.71802 -10.2997 < 2.2e-16 ***
## d:(Intercept) 984.88794 12.63336 77.9593 < 2.2e-16 ***
## e:(Intercept) 50.80033 7.87857 6.4479 6.468e-10 ***
## e:treatment2 42.64599 10.94359 3.8969 0.0001274 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
As a consequence of the specification ∼ treatment, the treatment coming

first in alphabetic order was taken as the reference level and then compared
to the other level, just as it would happen for a similar specification in lm().
This means that the summary output directly provides the three comparisons
between the two treatments (one for each model parameter). For the param-
eter b there is a difference between treatment of 0.14 (0.37), which is not
significantly different from 0 (p = 0.71), indicating that the slopes are similar
for the two treatments. For the parameters c and e (ED50) there are significant
differences: –358 (35; p < 0.0001) and 43 (11; p = 0.00013), respectively.
Thus, based on the difference in ED50 values only, we would conclude that
the mixture of phenmedipham and methyl oleate is less potent than phen-
medipham used alone. This conclusion is, however, somewhat contradicting
the picture in Figure 1.14 where it is clear that the mixture is more effective
as it reduces dry matter much more than phenmedipham alone, which does
not kill the plant completely even at the highest doses. The explanation is
that the comparison of ED50 values is relative to the estimated lower and
upper limits, and in this example the lower limits are quite different. So to
get the full picture, the lower limits also need to be considered (and perhaps
even the slopes). Simply relying on ED50 values (or any other EDxx values)
may be misleading. However, the interpretation also becomes more difficult as
differences in several model parameters have to be taken into consideration.
Occasionally interest lies in comparing the entire dose-response curve be-
tween treatments in one go, corresponding to looking at all three of the above-
mentioned comparisons. The approximate F -test offers a combined statistical
test for such a global comparison. It requires fitting a reduced model where a
single dose-response curve was fitted based on the pooled data from the two
herbicide treatments.
G.aparine.LL.4.pooled <- drm(drymatter ~ dose,

data = G.aparine,
fct = LL.4())
The F -test may now be invoked using the anova() method.

Continuous data 35
anova(G.aparine.LL.4.pooled, G.aparine.LL.4)
##
## 1st model
## fct: LL.4()
## pmodels: 1 (for all parameters)
## 2nd model
## fct: LL.4()
## pmodels: ~treatment, ~treatment, ~1, ~treatment
## ANOVA table
##
## ModelDf RSS Df F value p value
## 1st model 236 4927249
## 2nd model 233 2891677 3 54.673 0.000
As reported above, two out of three comparisons between the two treatments
of the individual model parameters gave significant results. Therefore, it is
not surprising that the global test is also significant (p < 0.0001): the p-value
from the global test may be thought of as a kind of weighted average of these
individual p-values.
1.2.2 Glyphosate and bentazone treatment of Sinapis alba

The potency of the two herbicides glyphosate and bentazone in white mus-
tard (Sinapis alba) was compared in an experiment reported by Christensen
et al. (2003). There were two independent sub-experiments for each herbi-
cide with the same seven doses plus an untreated control except for one
glyphosate sub-experiment where an eighth dose was included (also with four
replicates).
The experimental layout was planned as a completely randomized design
with four replicates for each herbicide dose except for the untreated con-
trol (dose 0) where 8 replicates were used. Unlike the previous example in
Subsection 1.2.1 the herbicide effects and the sub-experiment effects are not
confounded.
Data are provided in the dataset S.alba.comp.
head(S.alba.comp)
## exp herbicide dose drymatter Tf area Fo Fm

## 1 ben1 bentazone 0 4.1 200 31200 278 1662
## 2 ben1 bentazone 0 3.4 230 30600 278 1670
## 3 ben1 bentazone 0 2.6 210 27400 299 1646
## 4 ben1 bentazone 0 3.5 260 34600 288 1715
## 5 ben1 bentazone 0 4.3 200 31000 272 1651
## 6 ben1 bentazone 0 4.2 240 31400 286 1681
The first four columns are: exp denotes the sub-experiment (four levels),
herbicide denotes the herbicide applied, dose is the dose of the herbicide
applied (g a.i./ha), and drymatter is the response (in g/pot). There are four
more columns in the dataset, but they will not be used here.
We fit a joint model based on data from all four sub-experiments. Ide-
ally, we would fit a dose-response mixed-effects model that would treat the
sub-experiment variation as a separate source of variation. However, only hav-
ing two sub-experiments per herbicide treatment is not much and it may be
difficult to fit such a mixed model. We will come back to mixed models in
Chapter 7. Instead, we will look at two simpler approaches.
We may consider a joint model where we ignore the information about
the sub-experiments or, put in another way, pooling data from the two sub-
experiments for each herbicide. This approach corresponds to assuming that
the double number of replicates were used per dose (which is of course not
true). A joint model is convenient if we want to compare the two herbicides.
There is, however, no compelling reason for having a joint model encompassing
data from all sub-experiments. Moreover, it has been shown that, in theory,
estimated model parameters will be the same for a joint model and for separate
models per herbicide (Fang and Zhang, 2014). In practice, numerical issues
may lead to slight differences. Therefore, as an alternative, we may consider
separate models per sub-experiment and seek to combine them.
1.2.2.1 A joint dose-response model

We fit a joint model using drm(), but including the additional argument
curveid that provides information on the treatment groups in the data. For
each group a four-parameter log-logistic model may describe the average dose-
response relationship. It implies that a total of 4 · 2 = 8 parameters have to
be estimated.
S.alba.LL.4 <- drm(drymatter ~ dose,

curveid = herbicide,
data = S.alba.comp,
fct = LL.4())
If there had been a priori knowledge available about some of the model pa-
rameters being the same for both herbicides, then the argument pmodels could
have been used to incorporate such assumptions: Perhaps the lower and upper
limits for the two herbicides are identical, whereas slopes and ED50 param-
eters are different from herbicide to herbicide? It depends on what we know
about the experiment before collecting data.
The summary output shows that the lower and upper limits for the herbi-
cides are fairly similar, indicating that it may be meaningful to compare the
herbicides in terms of ED50 levels.
Continuous data 37
summary(S.alba.LL.4)
##
##
##
## b:bentazone 4.421705 1.376978 3.2112 0.001658 **
## b:glyphosate 1.606429 0.322669 4.9786 1.949e-06 ***
## c:bentazone 0.679894 0.092323 7.3643 1.673e-11 ***
## c:glyphosate 0.894469 0.138322 6.4666 1.756e-09 ***
## d:bentazone 4.018895 0.107033 37.5481 < 2.2e-16 ***
## d:glyphosate 3.874354 0.122103 31.7302 < 2.2e-16 ***
## e:bentazone 21.293475 1.348860 15.7863 < 2.2e-16 ***
## e:glyphosate 46.245932 6.379062 7.2496 3.074e-11 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
Figure 1.15 shows the fitted dose-response curves together with averages per
herbicide and dose. The model fit seems to be more or less appropriate.
The estimated relative potency based on the ED50s for the two herbicides
is obtained using the function EDcomp(), which takes the model fit as the
first argument and the ED levels (provided as percentages) to be compared
as the second argument. Additionally, we specify the argument interval =
"delta" to obtain confidence intervals based on the delta method.
EDcomp(S.alba.LL.4, c(50, 50), interval = "delta")
##
## Estimated ratios of effect doses
##
## Estimate Lower Upper
## bentazone/glyphosate:50/50 0.46044 0.32220 0.59868
Bentazone is approximately twice as potent as glyphosate. If the argument

interval = "delta" would be left out, then the p-value for testing the ratio
equal to 1 would be shown (it is very small).
We can also compare the two herbicides in terms of the difference in ED50
values using the function compParm(), which provides comparisons as differ-
ences or ratios. Here the difference is estimated by specifying "-" as the third
plot(S.alba.LL.4,
broken = TRUE,
ylab = "Dry matter (g/pot)")
bentazone
4 glyphosate
Dry matter (g/pot)
0 10 100 1000
Dose (g a.i./ha)
FIGURE 1.15
Four-parameter log-logistic model fitted to data on the effect of glyphosate
and bentazone on growth of white mustard (bentazone: solid line and open
circles, glyphosate: dashed line and triangles).
argument. The first two arguments are the model fit and the parameter of
interest.
compParm(S.alba.LL.4, "e", "-")
##
## Comparison of parameter 'e'
##
## bentazone-glyphosate -24.9525 6.5201 -3.827 0.0001988 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Bentazone has a significantly smaller ED50 as compared to glyphosate, but

also a significantly steeper slope as shown below using compParm().
Continuous data 39
compParm(S.alba.LL.4, "b", "-")
##
## Comparison of parameter 'b'
##
## bentazone-glyphosate 2.8153 1.4143 1.9906 0.04857 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Note that comparison of the same two parameters by means of a difference

and a ratio (such as the above two ED50 values) will usually not result in
the same p-value even though the same null hypothesis is being tested. The
reason is the use of the delta method to obtain the estimated standard error
for the ratio: a nonlinear change of scale of the parameter estimates will lead
to approximate standard errors (see also Section A.10).
1.2.2.2 Fitting separate dose-response models

The two-step approach proposed by Jiang and Kopp-Schneider (2014) is an
alternative to nonlinear mixed-effects models for dose-response analysis.
The first step consists of fitting a separate dose-response model for data
from each of the four independent sub-experiments. In principle the models
fitted to data from each independent sub-experiment need not be the same
as long as the same parameter of interest may be estimated from all models.
From each model fit, estimates of parameters of interest (with correspond-
ing estimated standard errors) are combined into a new dataset along with
information about treatments.
In a second step, these estimates are used as response values in a meta-
analytic random effects model where the experimental design is accounted
for through the use of fixed and random effects. In contrast to a usual linear
mixed model, the residual standard error is not estimated in the meta-analytic
random effects model. Instead, it is assumed to be non-constant and equal
to the estimated standard errors corresponding to the estimates. Inferential
procedures such as pairwise comparisons based on approximate Wald-type
U-tests are available, just like for an ordinary linear mixed model.
S.alba.LL.4.b1 <- drm(drymatter ~ dose,

data = subset(S.alba.comp, exp == "ben1"),
fct = LL.4())
S.alba.LL.4.b2 <- drm(drymatter ~ dose,

data = subset(S.alba.comp, exp == "ben2"),
fct = LL.4())
S.alba.LL.4.g1 <- drm(drymatter ~ dose,

data = subset(S.alba.comp, exp == "gly1"),
fct = LL.4())
S.alba.LL.4.g2 <- drm(drymatter ~ dose,

data = subset(S.alba.comp, exp == "gly2"),
fct = LL.4())
The next step is to extract the estimated ED50 values with the corresponding
standard errors for each sub-experiment: We define a 4×2 matrix of NAs. Then
we fill in row by row the estimated ED50 and the corresponding standard
errors (coef() and summary() and square brackets to extract the values in
row 4 and columns 1 and 2).
ed50.estimates <- matrix(NA, 4, 2)
ed50.estimates[1, ] <- coef(summary(S.alba.LL.4.b1))[4, 1:2]

ed50.estimates[2, ] <- coef(summary(S.alba.LL.4.b2))[4, 1:2]
ed50.estimates[3, ] <- coef(summary(S.alba.LL.4.g1))[4, 1:2]
ed50.estimates[4, ] <- coef(summary(S.alba.LL.4.g2))[4, 1:2]
ed50.estimates
## [,1] [,2]
## [1,] 29.26479 2.065501
## [2,] 18.85085 3.967966
## [3,] 25.56620 5.793800
## [4,] 62.73827 6.390966
Finally, we convert the matrix to a data frame with specified column names
(est and se) and a column with information about the treatments: the first
two rows contain estimates for the sub-experiments for bentazone and the last
two rows for glyphosate.
ed50.estimates <- as.data.frame(ed50.estimates)

names(ed50.estimates) <- c("est", "se")
ed50.estimates <- cbind(ed50.estimates,
treatment = c("bentazone", "bentazone",
"glyphosate", "glyphosate"))
ed50.estimates
## est se treatment
## 1 29.26479 2.065501 bentazone
## 2 18.85085 3.967966 bentazone
Continuous data 41
## 3 25.56620 5.793800 glyphosate

## 4 62.73827 6.390966 glyphosate
The estimates may be pooled into combined estimates using functionality from
the R package metafor (Viechtbauer, 2010). Specifically we use the function
rma(), which takes the estimated ED50 values as the first argument and the
corresponding squared standard errors as the second argument; the smaller
the standard error, the more weight will be assigned to the corresponding
estimate. The treatment variable is specified by means of a formula (as also
for many other model fitting functions) using the argument mods. Finally, the
argument data is used for providing the dataset where the variables given in
the first three arguments are found.
Specifically, we fit a weighted version of a one-way analysis of variance
model for ED50 (est) with squared standard errors of estimated ED50 values
as weights (se), using the argument mods = ~ treatment. These variables are
found in the dataset ed50estimates, which we defined above.
S.alba.ED50 <- rma(est, se^2, mods = ~ treatment,

data = ed50.estimates)
A condensed version of the summary output shows the estimated difference in

ED50 between the two herbicides: the estimated ED50 is 19.8 (18.7) larger for
glyphosate than for bentazone. This difference is not significant (p = 0.29).
round(coef(summary(S.alba.ED50)), 3)
## estimate se zval pval ci.lb ci.ub

## intrcpt 24.147 12.945 1.865 0.062 -1.224 49.518
## treatmentglyphosate 19.819 18.675 1.061 0.289 -16.784 56.421
Results from the joint and separate models disagree somewhat. The estimated
differences in ED50 are fairly similar. However, incorporating variation be-
tween sub-experiments, which is the more appropriate approach, makes it
more difficult to claim a significant difference as more variation in the data is
captured by this dose-response analysis.
2
Binary and binomial dose-response data
In this chapter, we enter the classical area of dose-response analysis, which

owes much to the seminal work by Finney (1971) as well as later contributions
by Morgan (1992) among others.
A binomial response is a sum of independent binary responses: 0/1, no/yes,
or absent/present; a binary response is a special case of a binomial response
(see also Subsection A.2.1). The total number of binary responses per concen-
tration or dose is decided upon as part of the experimental design and, hence,
it is known in advance, prior to observing the responses. These total numbers
need not be the same for all concentrations.
In some cases, the dose-response models for binomial data will be special
cases of generalized linear models for binomial data (e.g., experiments elic-
iting the full range from 0% to 100% response). However, in other cases the
dose-response models may be seen as extensions corresponding to generalized
linear models with link functions depending on one or more parameters (e.g.,
experiments where the control group will not show a 0% or 100% response).
Moreover, dose-response analysis of binomial data is based on statistical mod-
els that describe the mean trend on the probability scale, not on the log odds
scale as would typically be the case for a generalized linear model for binomial
data, e.g., logistic regression (McCullagh and Nelder, 1989, Chapter 4).
It is not unusual that the assumption of independence is not satisfied.
For instance, several organisms could share a container or several plants or
seeds could be placed in the same pot. Such departures may manifest them-
selves through so-called under- or, more commonly, over-dispersion where the
variation in the dose-response data is not adequately described by a model
relying on the assumption of independence. Consequently, the routine analysis
will need to be modified, either by attempting to augment the dose-response
model to allow extra-dispersion or, indirectly, by modifying standard errors
into robust ones but otherwise not changing the dose-response model. Which
approach to choose may to some extent depend on the data. Graphical model
checking by means of residuals is not very useful for binomial data and, thus,
model checking is more difficult. We will rely on visual assessment of plots of
fitted dose-response curves together with observed proportions.
library(drc)
library(devtools)
43
library(drcData)
library(multcomp)

2.1.1 Acute inhalation toxicity test
Racine et al. (1986) analyze data from a routine acute inhalation toxicity test
where 20 animals were exposed to 4 doses (mg/ml of an unknown substance),
5 animals per dose. The entire dataset is shown below.
acute.inh
## dose total num.dead

## 1 422 5 0
## 2 744 5 1
## 3 948 5 3
## 4 2069 5 5
We fit a two-parameter log-logistic model using the model fitting function
drm(). The proportions of dead animals (num.dead/total), which is the re-
sponse, are specified on the left-hand side of the tilde (∼). The independent
variable, which is denoted dose, is specified on the right-hand side of the
tilde. As we specify the response as proportions, we also need to give the
totals (total) through the argument weights. It informs the estimation pro-
cedure about the relative weighting of the binomial response values according
to the number of underlying binary responses (in case the response is binary
the weights would all be equal to 1 and they would not need to be speci-
fied). Next, we specify the dataset with the variables named num.dead, total,
and dose. The argument fct = LL.2() specifies a two-parameter log-logistic
model where the lower limit is fixed at 0 and the upper limit is fixed at 1.
Finally, the argument type indicates the type of response data to be analyzed.
This is different as compared to continuous dose-response data where it was
not necessary to specify this argument as the default is type = "continuous".
We store the resulting model fit in an object named acute.inh.LL.2:
acute.inh.LL.2 <- drm(num.dead/total ~ dose,
weights = total,
data = acute.inh,
fct = LL.2(),
type = "binomial")
Binary and binomial dose-response data 45
plot(acute.inh.LL.2, broken = TRUE,

xlim = c(0, 10000), ylim = c(0, 1),
ylab = "Proportion dead",
xlab = "Dose (mg/ml)")
1.0
0.8
Proportion dead
0.6
0.4
0.2
0.0
0 1000 10000
Dose (mg/ml)
FIGURE 2.1
Two-parameter log-logistic model fitted to the dataset acute.inh.
A plot of the original data and the fitted dose-response curve can give a visual
impression of how well the model fits the data, and with such a small dataset
it is possibly the only reasonable means of assessing the model. The plot is
shown in Figure 2.1. The model seems to provide a good fit to the data.
To see the parameter estimates we use the summary method.
summary(acute.inh.LL.2)
##
## limit at 0 and upper limit at 1 (2 parms)
##
##
## b:(Intercept) -7.9301 5.0812 -1.5607 0.1186
## e:(Intercept) 895.2982 83.5547 10.7151 <2e-16 ***

## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
From the output we can directly read off the estimated LD50 with the corre-
sponding standard error: 895 (84).
To obtain confidence intervals we can use the function confint(), which
by default will return (marginal) 95% confidence intervals.
confint(acute.inh.LL.2)
## 2.5 % 97.5 %
## b:(Intercept) -17.88909 2.028992
## e:(Intercept) 731.53397 1059.062485
The 95% confidence interval of the slope b extends below 0, indicating that
the dose-response data do not contain much information about the transition
between lower and upper limits. However, the 95% confidence interval of LD50
is well defined and it ranges from 732 to 1059. Both the estimated LD50 and
the 95% confidence intervals are in good agreement with the results obtained
by Racine et al. (1986), who used a different or more computationally involved
Bayesian approach.
2.1.1.1 Link to ordinary logistic regression

To see the link to ordinary logistic regression, we can fit exactly the same
model by means of the function glm():
acute.inh.glm <- glm(cbind(num.dead, total-num.dead) ~ log(dose),

data = subset(acute.inh, dose > 0),
family = binomial)
The resulting summary output looks like this:
coef(summary(acute.inh.glm))
## Estimate Std. Error z value Pr(>|z|)

## (Intercept) -53.90152 34.34623 -1.569358 0.1165646
## log(dose) 7.93001 5.08122 1.560651 0.1186062
As the parameterization used in glm() differs from the one used in drm(), the
parameter estimates will not all be the same: The estimated intercept in the
logistic regression model fit is equal to b·log(e) ((-7.9301) * log(895.2982)
= -53.90213). The estimated slope of log(dose) is identical to the estimated
slope parameter b in the above two-parameter log-logistic model fit except for
the change in sign. It is exactly the same model fit except for small numer-
ical differences due to different estimation procedures being used. For a re-
lated example using logistic regression, we refer to Venables and Ripley (2002,
pp. 190–194).
2.1.2 Tumor incidence

Shao (2012) analyzes binomial dose-response data describing liver tumor in-
cidence in female Sprague-Dawley rats that were exposed to the chemical
2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD). There are 6 non-zero concentra-
tions of TCDD and approximately 50 rats were exposed to each concentration.
The entire dataset is shown below.
liver.tumor
## conc total incidence

## 1 0.00 49 0
## 2 2.56 48 0
## 3 5.69 46 0
## 4 9.79 50 0
## 5 16.57 49 1
## 6 29.70 53 13
Fitting a two-parameter log-logistic model to the dataset liver.tumor may

be carried out using the following lines.
liver.tumor.LL.2 <- drm(incidence/total ~ conc,

weights = total,
data = liver.tumor,
fct = LL.2(),
type = "binomial")
The resulting summary output shows the estimated slope and EC50, where
EC50 may be interpreted as the concentration of TCDD resulting in a total
incidence of 50%:
summary(liver.tumor.LL.2)
##
##
##
## b:(Intercept) -4.9750 1.6897 -2.9443 0.003237 **
## e:(Intercept) 37.1774 4.1838 8.8859 < 2.2e-16 ***

## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
It may be much more relevant to estimate EC5 or EC10 to get quantities

that reflect realistic levels of exposure. Such EC values are so-called derived
parameters, which are functions of the model parameters, and they can be
estimated using the function ED().
ED(liver.tumor.LL.2, c(5, 10), interval = "delta")
##
##
## e:1:5 20.5705 2.5294 15.6129 25.5281
## e:1:10 23.9041 1.9778 20.0278 27.7804
The argument interval specifies the method applied for calculating confi-
dence intervals. Here we ask for intervals based on the delta method (Piegorsch
and Bailer, 2005, pp. 436–437), which are commonly reported, even though
they are approximate, only reaching exactly 95% coverage for large sample
sizes.
It is important to realize that by choosing the two-parameter log-logistic
model, the lower and upper asymptotes are a priori fixed at the values 0 and
1, respectively. It means that any response obtained for dose 0 is redundant
and will not be used in the estimation at all. For the same reason, large doses
will also only contribute little to the model fit. In other words, the choice of
model already contributes information about the dose-response relationship
and this piece of information cannot be updated based on the data as there
is no relevant parameter in LL.2() that could utilize such information (the
lower limit is fixed at 0). One way to see that dose 0 data add nothing to the
analysis is to repeat the analysis without dose 0:
liver.tumor.LL.2.no.zero <- drm(incidence/total ~ conc,
weights = total,
data = subset(liver.tumor, conc > 0),
fct = LL.2(),
type = "binomial")
Estimates and standard errors are exactly the same as for the model fitted to
the entire dataset:
summary(liver.tumor.LL.2.no.zero)
##
##
##
## b:(Intercept) -4.9750 1.6897 -2.9443 0.003237 **
## e:(Intercept) 37.1774 4.1838 8.8859 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
In contrast, removal of one or more of the non-zero concentrations where the

incidence is 0 will change the results (although not that much) as observations
for these concentrations are used for estimating the parameters b and e. We
revisit this data example in Subsection 6.1.3.
2.1.3 Earthworm toxicity test: Abbott’s formula

This subsection provides an example of dose-response analysis in case the data
exhibits natural mortality in the control group (dose 0). If there is any biolog-
ical or other reasons to expect natural mortality then it should be reflected in
the choice of dose-response model.
In contrast to the previous examples, the dose-response data in the present
example require an analysis that goes beyond the standard logistic regression
model. Because, by design of the experiment, the lower limit of the dose-
response curve cannot be assumed to be equal to 0 (the built-in assumption
for any standard logistic regression model). In the past, re-scaling of the re-
sponse, which is referred to as applying Abbott’s correction or formula, has
been a convenient, but an inappropriate way to deal with natural mortality
(e.g., Environment Canada, 2005, p. 46). It is convenient because it reduces
the subsequent analysis to fitting a (standard) two-parameter dose-response
model, like the ones fitted in the previous subsections. It is inappropriate be-
cause it introduces downwards or upwards bias (Hoekstra, 1987). Any such
pre-processing of data prior to fitting a dose-response model should be avoided;
this was already pointed out by Finney (1971, Chapter 7). Instead, a dose-
response model with a non-zero lower limit should be fitted.
Hoekstra (1987) analyze data from an earthworm toxicity test with 6
concentrations of the herbicide chloroacetamide. For each concentration, 40
earthworms were exposed to the herbicide and subsequently the number
of dead earthworms was counted. The control or natural mortality was
3/40 = 7.5%.
chlorac
## conc total num.dead

## 1 0 40 3
## 2 10 40 5
## 3 20 40 6
## 4 40 40 38
## 5 80 40 40
## 6 160 40 40
In order to incorporate natural mortality, we use a three-parameter dose-

response model where there is an additional model parameter describing the
proportion of natural mortality. Following Hoekstra (1987) we fit a three-
parameter log-normal model including a parameter for the lower limit (cor-
responding to the natural mortality), but fixing the upper limit at 1 as was
also the case for the two-parameter models in the previous examples. This is
an example of a generalized nonlinear model: a logistic regression model with
a parameter-dependent extension of the usual probit link function (Finney,
1971). The corresponding built-in function in drc is named LN.3u(), which is
short for LN.4(fixed = c(NA, NA, 1, NA)).
chlorac.LN.3u <- drm(num.dead/total ~ conc,

weights = total,
data = chlorac,
fct = LN.3u(),
type = "binomial")
The fitted dose-response curve captures the dose-response trend in the data
adequately, as shown in Figure 2.2.
Below, the summary output and the 95% confidence intervals are shown.
summary(chlorac.LN.3u)
##
## Model fitted: Log-normal with upper limit at 1 (3 parms)
##
##
## b:(Intercept) 4.603773 1.043813 4.4105 1.031e-05 ***
## c:(Intercept) 0.099988 0.033573 2.9783 0.002899 **
## e:(Intercept) 28.291922 2.271962 12.4526 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
confint(chlorac.LN.3u)
## 2.5 % 97.5 %
## b:(Intercept) 2.55793762 6.6496084
## c:(Intercept) 0.03418668 0.1657886
## e:(Intercept) 23.83895875 32.7448861
plot(chlorac.LN.3u, broken = TRUE,

xlim = c(0, 1000), ylim = c(0, 1),
xlab = "Concentration (mg/kg)")
abline(a = coef(chlorac.LN.3u)[2], b = 0, lty = 2)
1.0
0.8
Proportion dead
0.6
0.4
0.2
0.0
0 10 100 1000
Concentration (mg/kg)
FIGURE 2.2
Two-parameter log-logistic model fitted to the dataset chlorac. The dashed
line shows the estimated non-zero lower limit.
The estimated natural mortality, which is the parameter estimate for c con-
verted and rounded to an integer percentage, is equal to 10% with a 95%
confidence interval ranging from 3.4% to 17%. The estimated EC50 is equal
to 28.29 with a 95% confidence interval ranging from 24 to 33; Hoekstra (1987)
found a very similar result.
The EC50 corresponds to a 50% increase between the estimated lower
limit and the fixed upper limit of 1, that is the concentration resulting in
50% mortality beyond the natural mortality. It does not correspond to the
concentration resulting in a total mortality of 50%, which is an EC value that
is defined in absolute terms based on the probability scale and not relative
to limits partly or fully estimated from the data. For binomial data these
absolute EC values are often more relevant than the relative ones and they
may also be estimated using the function ED().
ED(chlorac.LN.3u, 50 /100, type = "absolute")
##
##
## e:1:0.5 27.4464 2.3155
If instead a two-parameter model (e.g., LN.2()) had been fitted, then the
estimated EC50 would become smaller, biased downwards, and it would have
narrower 95% confidence intervals. In short, a less accurate but more precise
estimate of EC50 would be the result; this is an example of the so-called
bias-variance tradeoff occurring when choosing between different models.
2.1.4 Another earthworms toxicity test: Estimating the

upper limit
The dataset earthworms contains the number of earthworms remaining in a
container that is contaminated with a toxic substance (not disclosed) instead
of migrating to the neighbouring uncontaminated container.
For dose 0, a fifty-fifty distribution between the two containers is to be
expected. Therefore, we fit a three-parameter log-logistic model to binomial
data. This is another example of a generalized nonlinear model; this time it
is a logistic regression model with a parameter-dependent logit link function.
This model is fitted as follows in R.
earthworms.LL.3 <- drm(number/total ~ dose,
weights = total,
data = earthworms,
fct = LL.3(),
type = "binomial")
summary(earthworms.LL.3)
##
##
##
## b:(Intercept) 1.505679 0.338992 4.4416 8.928e-06 ***
## d:(Intercept) 0.604929 0.085800 7.0505 1.783e-12 ***
## e:(Intercept) 0.292428 0.083895 3.4856 0.000491 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Perhaps even better (truer to the experimental design) would be to fit a log-
logistic model where the upper limit is not estimated but instead fixed at the
value 0.5. This is achieved using the following R lines:
earthworms.LL.3.fixed <- drm(number/total ~ dose,

weights = total,
data = earthworms,
fct = LL.3(fixed = c(NA, 0.5, NA)),
type = "binomial")
summary(earthworms.LL.3.fixed)
##
##
##
## b:(Intercept) 1.646689 0.376494 4.3737 1.221e-05 ***
## e:(Intercept) 0.377269 0.076785 4.9133 8.956e-07 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
By fixing the upper limit, there is a very slight gain in precision for the esti-
mated ED50 (the parameter e), as the standard error becomes a little smaller.
In contrast, but in this case of less interest, the precision of the slope param-
eter b is reduced as the dose 0 is important for estimating the slope. Note
that the above comparison between model fits is meant to show how different
model choices affect results. However, we want to stress that in practice, ide-
ally, a single model should be chosen up front based on what is known about
the experimental design. If this is not feasible, then model averaging is an
alternative (see Chapter 6 for more details and examples).

In this section, we consider two data examples with two and four dose-response
curves, respectively. An entire dose-response curve corresponds to a treatment.
Ideally, such grouped dose-response data would be analyzed using mixed-
effects dose-response models or related techniques. In some cases it may, how-
ever, not be feasible to apply such complex methodology.
In both examples below, the main interest lies in comparing treatments

in terms of some features of the dose-response curves, which may be cap-
tured by model parameters such as LD50 values. In the first example, there
are replicated dose-response sub-experiments per treatment. The second ex-
ample lacks replicated dose-response curves as only one dose-response curve
corresponds to one treatment. In such a case there is no way (based on the
data alone) to separate random variation between dose-response curves (due
to various unobserved differences in the execution of each dose-response sub-
experiment) from systematic differences between treatments. Consequently,
only weak claims on treatment differences may be made based on such an
experimental design.
2.2.1 Toxicity of fluoranthene under different ultraviolet

radiation
Wheeler et al. (2006) report data from an experiment in which fathead min-
now larvae (Pimephales promelas) were exposed to fluoranthene, a polycyclic
aromatic hydrocarbon, under two different algal densities resulting in differ-
ent levels of ambient ultraviolet radiation. The two algal concentrations were
0.7 and 1.5 times 10,000 cells. For each algal concentration, four groups of
larvae were exposed to fluoranthene concentrations of 5, 10, 20, and 30 µg/L
for 4 hours. The number of dead fish larvae out of the total number of lar-
vae was the binomial response of interest. This dose-response experiment was
repeated three times (no information for identifying individual experiments
was provided). Therefore, we cannot model the experiment-level variation di-
rectly in the dose-response model. For now, we note that the assumption of
independent binomial responses might be violated.
head(fluoranthene)
## algalconc conc mortality totalnum

## 1 0.7 5 0 23
## 2 0.7 5 0 19
## 3 0.7 5 0 20
## 4 1.5 5 0 22
## 5 1.5 5 0 25
## 6 1.5 5 0 24
We fit a two-parameter log-normal (“log-probit”) model to the data although

a probit model was originally fitted by Wheeler et al. (2006). We prefer the
log-normal model as it is only defined for non-negative doses.
plot(fluoranthene.LN.2.1, broken = TRUE, type = "all",

xlim = c(0, 100), ylim = c(0, 1),
xlab = expression(paste("Concentration (", mu, "g/L)")),
legendPos = c(5, 1))
1.0
0.7
0.8 1.5
Proportion dead
0.6
0.4
0.2
0.0
0 10 100
Concentration (µg/L)
FIGURE 2.3
Two-parameter log-normal model fitted to the dataset fluoranthene, one
curve for each algal concentration.
fluoranthene.LN.2.1 <- drm(mortality/totalnum ~ conc,

curveid = algalconc,
weights = totalnum,
data = fluoranthene,
fct = LN.2(),
type = "binomial")
Note that the second argument in the model specification identifies the in-
dividual dose-response curves in the data and here we can only identify the
two treatment groups, but not the underlying individual dose-response ex-
periments. Figure 2.3 shows that the fitted dose-response curves capture the
trends in the data.
Parameter estimates and the corresponding 95% confidence intervals are

shown below.
summary(fluoranthene.LN.2.1)
##
## Model fitted: Log-normal with lower limit at 0 and upper
##
##
## b:0.7 2.70147 0.28738 9.4004 < 2.2e-16 ***
## b:1.5 2.87566 0.31085 9.2511 < 2.2e-16 ***
## e:0.7 15.10005 0.67387 22.4080 < 2.2e-16 ***
## e:1.5 17.90466 0.75721 23.6456 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
confint(fluoranthene.LN.2.1)
## 2.5 % 97.5 %
## b:0.7 2.138217 3.264721
## b:1.5 2.266412 3.484906
## e:0.7 13.779295 16.420807
## e:1.5 16.420556 19.388755
From the summary it seems that the two slopes could be assumed to be iden-
tical, an assumption about parallelism. However, ideally such an assumption
should have been imposed before fitting any dose-response model to avoid
inflation of standard errors.
As the confidence interval for the two LC50 values are barely overlapping,
it could be tempting to claim that the p-value for testing that the two LC50
values are the same is just above 0.05, almost significant. However, in reality
the two LC50 values are clearly significantly different. This conclusion may
be reached from the direct comparison between the two LC50 values. We use
the function compParm() for making comparisons between treatments for a
specific model parameter:
compParm(fluoranthene.LN.2.1, "e", "-")
##
##
## 0.7-1.5 -2.8046 1.0136 -2.7669 0.00566 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
where "e" indicates the model parameter of interest and "-" specifies that
comparisons should be in terms of differences (the other option being ratios).
The two LC50 values are clearly significantly different.
In an attempt to incorporate the experiment-level variation (as well as
any other model misspecification) into the dose-response analysis, we may
use the sandwich variance estimator for deriving robust standard errors (see
Section A.5 for more details).
compParm(fluoranthene.LN.2.1, "e", "-", vcov = sandwich)
##
##
## 0.7-1.5 -2.80460 0.97744 -2.8693 0.004113 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
The conclusion remains the same: there is a clearly significant difference be-
tween the two groups in terms of LC50 values (estimated difference: –2.8
(0.977) µg/L, p = 0.004).
2.2.2 Toxicity of different types of selenium

In this example, we consider the comparison of four dose-response curves
corresponding to four types of selenium (Jeske et al., 2009). Varying numbers
of concentrations (5–8, unit unknown) and partly different concentrations of
selenium were used. The number of dead flies was recorded. The total number
of flies ranged from 63 to 188.
A summary of the dataset is shown below.
summary(selenium)
## type conc total dead

## Min. :1.00 Min. : 0.0 Min. : 63.0 Min. : 0.0
## 1st Qu.:2.00 1st Qu.: 25.0 1st Qu.:139.0 1st Qu.: 11.0
## Median :3.00 Median :100.0 Median :146.0 Median : 40.0
## Mean :2.48 Mean :190.4 Mean :142.2 Mean : 54.4
## 3rd Qu.:3.00 3rd Qu.:300.0 3rd Qu.:153.0 3rd Qu.: 85.0
## Max. :4.00 Max. :800.0 Max. :188.0 Max. :188.0
First, we fit a two-parameter log-logistic model assuming both different slopes

and LC50 values for the different types of selenium.
selenium.LL.2.1 <- drm(dead/total ~ conc,

curveid = type,
weights = total,
data = selenium,
fct = LL.2(),
type = "binomial")
plot(selenium.LL.2.1, type = "all", broken = TRUE,

xlim = c(0, 10000),
xlab = "Concentration",
legendPos = c(5, 0.95))
The plot of the data with the fitted dose-response curves superimposed shows
that the fitted dose-response curves capture the trends in the data quite well
(Figure 2.4).
1.0
1
0.8
Proportion dead
2
3
0.6
4
0.4
0.2
0.0
0 10 100 1000 10000
Concentration
FIGURE 2.4
Two-parameter log-logistic model fitted to the dataset selenium. The different
line types correspond to the different types of selenium.
The argument legendPos was used to specify the position of the legend
(more precisely: the top right corner of the box containing the legend).
Next, we fit a two-parameter log-logistic model, but now assuming different
slopes and a common LC50 parameter for all four types of selenium.
selenium.LL.2.2 <- drm(dead/total~conc,

curveid = type,
weights = total,
data = selenium,
fct = LL.2(),
type = "binomial",
pmodels = list(~factor(type) - 1, ~1))
Using the usual formula specification in R, we use the argument pmodels

to specify that there should be different slope parameters b for the different
treatments (∼factor(type) - 1) and a common LC50 for all four treatments
(∼1).
We can now compare the two model fits using a likelihood ratio test (an
approximate chi-square test) by means of the function anova(), just like for
linear or generalized linear models in R.
anova(selenium.LL.2.2, selenium.LL.2.1)
##
## 1st model
## fct: LL.2()
## pmodels: ~factor(type) - 1, ~1
## 2nd model
## fct: LL.2()
## pmodels: type (for all parameters)
## ANOVA-like table
##
## ModelDf Loglik Df LR value p value
## 1st model 5 -437.99
## 2nd model 8 -376.21 3 123.56 0
We can safely conclude that the four LC50 values are not identical (p <
0.0001). Jeske et al. (2009) reached the same conclusion. The next step is
to identify differences between the different types of selenium. The function
EDcomp() provides all comparisons of LC50 values in terms of ratios:
EDcomp(selenium.LL.2.1, c(50, 50))
##
##

## 1/2:50/50 6.6653e-01 7.8374e-02 -4.2548e+00 2.0922e-05
## 1/3:50/50 2.1072e+00 1.5539e-01 7.1253e+00 1.0387e-12
## 1/4:50/50 2.8405e+00 3.1654e-01 5.8146e+00 6.0771e-09
## 2/3:50/50 3.1614e+00 3.6398e-01 5.9383e+00 2.8804e-09
## 2/4:50/50 4.2617e+00 6.0623e-01 5.3803e+00 7.4368e-08
## 3/4:50/50 1.3480e+00 1.4672e-01 2.3721e+00 1.7689e-02
All four types of selenium are significantly different from one another (highest
p-value is 0.018). For all types of selenium we calculate both marginal con-
fidence intervals where each single interval has 95% coverage (but this also
means that the collection of confidence interval does not have simultaneous
95% coverage) as well as adjusted 95% confidence intervals (with adjustment
for simultaneous inference such that the collection of confidence intervals has a
95% coverage). Marginal confidence intervals are obtained using the function
ED().
ED(selenium.LL.2.1, c(50), interval = "delta")
##
##
## e:1:50 252.2556 13.8268 225.1555 279.3556
## e:2:50 378.4605 39.3707 301.2953 455.6256
## e:3:50 119.7132 5.9054 108.1389 131.2875
## e:4:50 88.8053 8.6161 71.9180 105.6926
To obtain adjusted confidence intervals, we also initially use the function ED()
but now supplying the argument multcomp = TRUE to augment the output
with a component named EDmultcomp that may be supplied directly to the
function glht() in the package multcomp (Hothorn et al., 2008). Further-
more, the argument display = FALSE suppresses the printing of marginal
confidence intervals, which we have already obtained above.
selenium.EDres <- ED(selenium.LL.2.1, c(50),
interval = "delta",
multcomp = TRUE,
display = FALSE)
Now we can apply glht() to the object named selenium.EDres. However,

to obtain adjusted confidence intervals we can additionally apply confint()
directly to the result from glht().
confint(glht(selenium.EDres[["EDmultcomp"]]))
##
## Simultaneous Confidence Intervals
##
## Fit: NULL
##
## Quantile = 2.4908
## 95% family-wise confidence level
##
##
## Linear Hypotheses:
## Estimate lwr upr
## e:1:50 == 0 252.2556 217.8151 286.6960
## e:2:50 == 0 378.4605 280.3942 476.5267
## e:3:50 == 0 119.7132 105.0039 134.4225
## e:4:50 == 0 88.8053 67.3438 110.2667
The resulting adjusted confidence intervals become somewhat wider as com-

pared to the unadjusted marginal ones because of the multiplicity adjustment.
In this case the adjustment corresponds to the Bonferroni adjustment as the
correlations between estimates are all equal to 0 (Pipper et al., 2012).
It is also possible to obtain all pairwise comparisons of LC50 between
the four different types of selenium. Again, we utilize the functionality of
multcomp: the function call contrMat(1:4, "Tukey") constructs a matrix of
all pairwise comparisons of four groups.
summary(glht(selenium.EDres[["EDmultcomp"]],
linfct = contrMat(1:4, "Tukey")))
##
## Simultaneous Tests for General Linear Hypotheses
##
## Multiple Comparisons of Means: Tukey Contrasts
##
##
## Estimate Std. Error z value Pr(>|z|)
## 2 - 1 == 0 126.20 41.73 3.024 0.0111 *
## 3 - 1 == 0 -132.54 15.04 -8.816 <0.001 ***
## 4 - 1 == 0 -163.45 16.29 -10.033 <0.001 ***
## 3 - 2 == 0 -258.75 39.81 -6.499 <0.001 ***
## 4 - 2 == 0 -289.66 40.30 -7.187 <0.001 ***
## 4 - 3 == 0 -30.91 10.45 -2.959 0.0131 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## (Adjusted p values reported -- single-step method)
Using this approach, arbitrary LCxx values may be compared between groups.
Moreover, it also works for other types of dose-response data.
3
Count dose-response data
Count data include number of fronds, offspring, juveniles, leaves, or roots, i.e.,
non-negative integers. In contrast to binomial data, the experimental design
imposes no a priori upper limit on the counts. In theory, there may be no
upper limit although very large counts will be very unlikely. In practice, a
limited number of different counts may be observed, but some counts may
occur multiple times (with ties). Ideally, counts should be recorded over the
same time period in order to be comparable. However, there are ways to adjust
for varying durations of the time period. For instance, reproduction data,
which are commonly obtained from chronic toxicity tests in ecotoxicology, are
often counts of the number of offspring present at the end of the test period
(a certain pre-specified time period).
Dose-response analysis of count data is based on statistical models that
describe the mean trend on the scale of the counts, not on the scale of the
logarithm-transformed mean as would typically be the case for a generalized
linear model for count data (McCullagh and Nelder, 1989, Chapter 6). How-
ever, the same distributions as used in generalized linear models for counts
may be used when fitting dose-response models.
There may be several ways to analyze such count data, depending on the
distributional and modeling assumptions one is willing to make (Ritz and Van
der Vliet, 2009). In general the closer the chosen distributional assumptions
are to the true distributions, which generated the data, the more efficient (i.e.,
smaller standard errors) the statistical analysis will be. In practice we may
not get close to the true distributions, and it will often be a trade-off between
more bias in parameter estimates if we are being too confident and choose a
quite specific model, and more imprecision (larger estimated standard errors)
if we are less picky and choose a model that does not seem fully appropriate
(a bias-variance trade-off).
Counts are often assumed to follow a Poisson distribution, although this
distribution is not always sufficiently flexible to describe the variation in the
count dose-response data (see also Subsection A.2.2.1). The standard devia-
tion of Poisson distributed counts is completely determined by the mean of
the distribution; there is no separate parameter for the standard deviation as
is the case of the normal distribution. Therefore, it may easily happen that a
Poisson distribution does not adequately capture the variation in the observed
counts: the counts may exhibit less or more variation than predicted by the
Poisson distribution. Excess variation is usually referred to as over-dispersion
63
(e.g., Morgan, 1992, Chapter 6), just as for binomial data. Nevertheless, a
Poisson dose-response model may be a good starting point for the analysis.
Apart from the distributional assumptions, dose-response analysis is car-
ried out in much the same way as for continuous dose-response data (Chap-
ter 1), e.g., model checking by means of residuals.
library(drc)
library(devtools)
library(drcData)
library(ggplot2)
library(lmtest)
library(multcomp)
library(dplyr)
library(sandwich)

3.1.1 Counting number of fronds
Ritz and Van der Vliet (2009) report dose-response data from a sublethal
Lemna minor toxicity test where gradual reduction in growth may be ob-
served. Thus, more information may be elicited from this test than from a
toxicity test where a binary response is observed (see Chapter 2). Specifi-
cally, numbers of fronds from Lemna minor, which is an aquatic freshwater
plant (common duckweed), were counted for a range of concentrations, which
were different dilutions of a metal mining effluent. The dilutions (in percent
of the effluent, % v/v) were chosen to ensure that both partial effects and
strong inhibitory effects would be observed such that the entire dose-response
relationship could be described.
In total 44 plants were examined, corresponding to 10 concentrations in-
cluding the control. There were 4 replicates for all concentrations except for
the control where 8 replicates were used. As seen below, the observed num-
bers of fronds range from 61–70 in the control group to 29–34 at the highest
concentration.
Count dose-response data 65
head(lemna)
## conc frond.num
## 1 0 70
## 2 0 66
## 3 0 61
## 4 0 65
## 5 0 65
## 6 0 61
lemna %>%
group_by(conc) %>%
summarize(min = min(frond.num),
max = max(frond.num))
## # A tibble: 10 x 3
## conc min max
## <dbl> <dbl> <dbl>
## 1 0 61 70
## 2 0.38 64 67
## 3 0.76 54 65
## 4 1.52 55 58
## 5 3.03 50 63
## 6 6.06 49 56
## 7 12.1 41 47
## 8 24.2 37 37
## 9 48.5 31 36
## 10 97 29 34
We will fit a Poisson dose-response model, although in our experience, it is

not often that a Poisson distribution provides a good description of dose-
response count data. In the subsequent subsections, we show how to improve
the dose-response model. We will assume that a three-parameter log-logistic
model (LL.3()) describes the mean trend in the data adequately.
lemna.minor.LL.3 <- drm(frond.num ~ conc,

data = lemna,
fct = LL.3(),
type = "Poisson")
It is possible to some extent to validate model assumptions by means of the

usual graphical model diagnostics. Below we show a residual plot (using raw
residuals will suffice):
plot(fitted(lemna.minor.LL.3), residuals(lemna.minor.LL.3),
xlab = "Fitted values",
ylab = "Raw residuals")
abline(h=0, lty=2)
Raw residuals
5
0
−5
30 40 50 60
Fitted values
FIGURE 3.1
The residual plot for the three-parameter log-logistic model fitted to the
Lemna minor data.
Based on the residual plot in Figure 3.1 we can conclude that there are no
substantial departures from random scatter (around the y axis). Therefore,
it seems reasonable to assume that a three-parameter log-logistic model will
suffice. We do not look at the standard QQ normal probability plot as it is
not helpful in assessing whether or not the data are Poisson distributed.
The plot of the data with the fitted dose-response curve superimposed is
shown below. Figure 3.2 also supports our initial impression that the three-
parameter log-logistic model describes the data adequately. However, it is
clear that the data hold almost no information about what would happen
for very large concentrations. The assumption about a lower limit of 0 is
crucial as it means that one piece of additional information (not found in the
data!) is incorporated into the model. This consideration is analogous to what
we discussed for continuous dose-response data in Subsection 1.1.5. In case
there are no strong biological reasons for a particular assumption, it may be
plot(lemna.minor.LL.3, type = "all", broken = TRUE,

xlab = "Effluent concentration (% v/v)",
ylab = "Number of fronds",
xlim = c(0, 500), ylim = c(0,72),
conName = "Control", lwd = 2)
70
60
Number of fronds
50
40
30
20
10
0
Control 1 10 100
Effluent concentration (% v/v)
FIGURE 3.2
The fitted three-parameter log-logistic dose-response curve together with the
Lemna minor data.
advisable to fit multiple models such as three- and four-parameter models in

order to appreciate the effect of the absence or presence of such an additional
assumption.
Once we have accepted the model fit as adequately describing the dose-
response data, we can proceed to obtain the estimate of the parameter(s) of
interest, which in this case is a single parameter, namely EC50. It is impor-
tant to realize that the interpretation of EC50 and indeed any other effective
concentration hinges on the chosen dose-response model function. For the
three-parameter log-logistic model function LL.3(), the estimated EC50 is
the concentration yielding a 50% reduction of the average number of fronds
relative to the average level for the control group, which is the upper limit of
the fitted dose-response curve (corresponding to the estimate of the parameter
d). This definition implicitly assumes that the lower limit of the fitted dose-
response is fixed at 0, corresponding to the plausible biological assumption
that sufficiently high concentrations will result in no growth at all, namely
zero counts. Assuming a non-zero lower limit (LL.4()) may result in a some-
what different result because the estimated EC50 is interpreted relative to
the estimated lower and upper limits. In this case, the estimated EC50 be-
comes approx. 5 times smaller as shown by Van der Vliet and Ritz (2013).
This point also applies to continuous dose-response data, but not to binomial
dose-response data.
summary(lemna.minor.LL.3)
##
##
##
## b:(Intercept) 0.49207 0.07418 6.6335 3.278e-11 ***
## d:(Intercept) 66.79414 2.59857 25.7042 < 2.2e-16 ***
## e:(Intercept) 56.07520 14.27537 3.9281 8.562e-05 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
The estimated EC50 is 56.1 with a corresponding standard error of 14.3. Thus,
the average number of fronds is reduced by 50% relative to an average level
of 66.8, i.e., reaching an average of 33.4, at a concentration of 56.1 (effluent
% v/v).
As seen previously for continuous data, the generic way to extract EC50
would be to use the function ED(), which may also be used for obtaining other
EC values:
ED(lemna.minor.LL.3, c(10, 20, 50))
##
##
## e:1:10 0.64497 0.46796
## e:1:20 3.35159 1.67765
## e:1:50 56.07520 14.27537
The standard errors of the estimated EC10 and also EC20 to some extent
are quite large and the corresponding Wald-type 95% confidence interval will
have an unrealistic negative lower limit.
ED(lemna.minor.LL.3, c(10, 20, 50), interval = "delta")
##
##
## e:1:10 0.644967 0.467963 -0.272224 1.562158
## e:1:20 3.351589 1.677654 0.063448 6.639731
## e:1:50 56.075199 14.275369 28.095989 84.054408
In principle, it would be possible to reparameterize the model function in

such a way that the 95% confidence interval of EC10 is fully contained in
the non-negative concentration scale, with a positive lower limit. However,
this manoeuvre only conceals the fact that the chosen dose-response model is
not capable of eliciting much information on EC10 from the data (possibly
because there is almost no information in the data to estimate EC10). Another,
possibly improved model may or may not be able to do a better job.
3.1.2 Counting offspring: Modeling hormesis

Bailer and Oris (1997) analyzed data from an experiment where the number
of offspring of C. dubia, which is a species of water flea used in toxicity testing,
was counted after being exposed to different concentrations of waste water for
7 days.
Again, we start out looking at the data. The table below shows that there
were 10 replicates for each of five concentrations, one of which is a control
group. Next, we can plot the raw data:
head(C.dubia)
## conc number
## 1 0 27
## 2 0 30
## 3 0 29
## 4 0 31
## 5 0 16
## 6 0 15
with(C.dubia, table(conc))
## conc
## 0 1.56 3.12 6.25 12.5
## 10 10 10 10 10
Bailer and Oris (1997) fitted a generalized linear model assuming a Poisson
distribution and including both linear and quadratic terms of concentrations
in order to describe the hormetic effect, which may be inferred from the ini-
tial increase in the response for low concentrations, followed by a decreasing
response for increasing concentrations (Figure 3.3).
plot(number ~ conc, data = C.dubia,

xlab = "Waste water concentration (%)",
ylab = "Number of offspring",
ylim = c(0, 50))
50
Number of offspring
40
30
20
10
0
0 2 4 6 8 10 12
Waste water concentration (%)
FIGURE 3.3
The scatter plot of the number of offspring as a function of waste water con-
centration (%) for the C. dubia data.
We take a different route by considering dose-response models as defined

within the framework of this book. Specifically, we start out fitting a three-
parameter log-logistic model assuming that the counts are Poisson distributed
(inspired by the previous example); this is an example of an extended general-
ized linear model where the link function depends on an unknown parameter
(in this case the upper limit d).
C.dubia.LL.3 <- drm(number ~ conc,

data = C.dubia,
fct = LL.3(),
type = "Poisson")
The model fit shown in Figure 3.4 (top panel) is not good. For instance, the
predicted mean offspring at concentration 0 is not centered among the data
points for concentration 0: it has been shifted upwards because of the hormetic
effect observed for higher concentrations. Perhaps even more pronounced: the
predicted mean number of offspring is too small for the low concentrations
that seemed to produce a strong hormetic effect.
As a second attempt we consider the four-parameter Brain-Cousens horme-
sis model, which may be specified using the function BC.4(). This means we
retain the assumption that the lower limit is equal to 0:
C.dubia.BC.4 <- drm(number ~ conc,

data = C.dubia,
fct = BC.4(),
type = "Poisson")
The fitted dose-response curve is shown in Figure 3.4 (bottom panel). This
time the fitted dose-response curve describes the trend in the data very closely
(too closely?).
plot(C.dubia.LL.3, type = "all", broken = TRUE,

xlim = c(0, 100),
ylim = c(0, 50))
plot(C.dubia.BC.4, type = "all", broken = TRUE,

xlim = c(0, 100),
ylim = c(0, 50))
Once we have found an appropriate dose-response model, we can take a closer

look at the parameter estimates:
summary(C.dubia.BC.4)
##
## Model fitted: Brain-Cousens (hormesis) with lower limit fixed
## at 0 (4 parms)
##
##
## b:(Intercept) 3.83037 0.39251 9.7585 < 2.2e-16 ***
## d:(Intercept) 21.80069 1.38689 15.7191 < 2.2e-16 ***
## e:(Intercept) 7.60327 0.61601 12.3427 < 2.2e-16 ***
## f:(Intercept) 4.03279 0.88706 4.5462 5.462e-06 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
50
Number of offspring 40
30
20
10
0 1 10 100
50
Number of offspring
40
30
20
10
0 1 10 100
FIGURE 3.4
The fitted three-parameter log-logistic and four-parameter Brain-Cousens
hormesis models shown together with the C. dubia data.
The parameter f quantifies the extent of the hormetic effect in the sense
that the more positive this parameter is, the larger is the hormesis peak.
The parameter f is significantly different from 0, implying that there is a
substantial hormetic effect. Bailer and Oris (1997) reached the same conclusion
based on a quadratic generalized linear model.
The summary output does not provide any estimate of EC50. In contrast
to the four-parameter log-logistic model, the parameter e has lost its interpre-
tation as EC50 in the hormesis models. Therefore, to estimate EC50, we use
the function ED():
ED(C.dubia.BC.4, 50)
##
##
## e:1:50 11.7863 0.5086
For hormesis models and biphasic dose-response models in general, it may be
possible to estimate EC values for both phases, i.e., corresponding to effects
to the left and right of the peak. For instance, we can estimate EC-10 and
EC-20 as follows (notice the use of the additional argument bound = FALSE,
which switches off checking if the specified effect levels lie between 0 and 100):
ED(C.dubia.BC.4, c(-10, -20), bound = FALSE)
##
##
## e:1:-10 8.09294 0.46267
## e:1:-20 7.61867 0.49695
Actually, EC-10 and EC-20 turn out to be almost as precisely determined from
the data as is EC50 (with comparable standard errors), reflecting that these
estimates are within the concentration range where actual concentrations are
found in the data.
3.1.3 More counting offspring: Varying observation periods

Delignette-Muller et al. (2014b) reported reproduction data for Daphnia
Magna exposed to different concentrations of the insecticide chlordane (µg/L).
There are 6 concentrations, including one control group, and 10 replicates per
concentration. Specifically, numbers of offspring were counted over an obser-
vation period of 21 days. If a daphnia died during the 21, days the number of
offspring until death was counted. For daphnias, the duration of the observa-
tion period was observed. We start out by looking at the data.
head(chlordan)
## replicate conc repro time

## 1 1 0.00 125 21.0
## 2 1 0.18 89 21.0
## 3 1 0.73 90 21.0
## 4 1 1.82 42 21.0
## 5 1 2.90 29 21.0
## 6 1 7.00 10 11.5
As a first naı̈ve approach we can fit a three-parameter log-logistic model, as-
suming Poisson distributed counts but ignoring the duration of the observation
periods. This model describes the total number of offspring over 21 days.
chlordan.LL.3 <- drm(repro ~ conc,
data = chlordan,
fct = LL.3(),
type = "Poisson")
The residual plot shown in Figure 3.5 looks reasonable. Likewise, the fit-
ted dose-response model seems to describe the trend in the data adequately
(Figure 3.6).
Now we can take a look at the summary output:
summary(chlordan.LL.3)
##
##
##
## b:(Intercept) 1.089500 0.070639 15.423 < 2.2e-16 ***
## d:(Intercept) 113.666336 2.984324 38.088 < 2.2e-16 ***
## e:(Intercept) 1.615563 0.127838 12.638 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
The estimated EC50 is 1.62 (0.13). The estimated average total reproduction
over 21 days in the control group is 114 (2.98) offspring.
Using the confint method we find the 95% confidence interval for EC50.
confint(chlordan.LL.3)
## 2.5 % 97.5 %
## b:(Intercept) 0.9510493 1.22795
## d:(Intercept) 107.8171683 119.51550
## e:(Intercept) 1.3650065 1.86612
plot(fitted(chlordan.LL.3), residuals(chlordan.LL.3))
abline(h=0, lty=2)
40
residuals(chlordan.LL.3)
20
0
−20
−60
20 40 60 80 100
fitted(chlordan.LL.3)
FIGURE 3.5
The residual plot for the three-parameter log-logistic model fitted to the
chlordan dataset.
As already mentioned, there is some imbalance in durations of observation

periods, which implies that counts even obtained at the same concentration
may not be comparable as they were recorded over varying time periods.
Specifically, the distribution of durations of observation periods looks like this:
with(chlordan, table(time))
## time
## 2.5 9.5 11.5 14.5 17.5 18.5 20.5 21
## 1 2 3 1 1 9 1 42
The above dose-response model did not incorporate this imbalance and, hence,
it may be seen as a misspecified model (as the distributional assumptions are
partly misspecified). Ideally, such imbalance should be taken into account in
the statistical analysis in order to reduce bias in the estimates.
One approach is to fit a three-parameter log-logistic model but use the du-
rations of observation periods as weights; this means that numbers of offspring
plot(chlordan.LL.3,
broken = TRUE,
type = "all",
xlim = c(0, 10),
xlab = "Concentration (mu g/L)",
ylab = "Number of offspring")
120
Number of offspring
100
80
60
40
20
0
0 0.1 1 10
Concentration (mu g/L)
FIGURE 3.6
The fitted three-parameter log-logistic dose-response curve to the chlordan
dataset shown together with the raw data.
are scaled by their durations. This again means that the average number of
offspring per day for a given concentration is modeled. In particular, the pa-
rameter d can be interpreted as the average number of offspring per day in the
unexposed control group. However, it is important to stress that the scaling
or normalization by duration is taken care of by the model.
The weighted Poisson model is fitted as follows using the argument
weights:
chlordan.LL.3.we <- drm(repro ~ conc,

data = chlordan,
fct = LL.3(),
type = "Poisson",
weights = time)
The resulting summary output is obtained here:

summary(chlordan.LL.3.we)
##
##
##
## b:(Intercept) 0.929859 0.065783 14.135 < 2.2e-16 ***
## d:(Intercept) 5.544291 0.146147 37.936 < 2.2e-16 ***
## e:(Intercept) 1.758823 0.151862 11.582 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
The estimated EC50 is now 1.76 (0.15). The estimated average daily repro-
duction rate in the control group is 5.54 (0.146) offspring.
Looking at Figure 3.6 (note in particular the variation in the mid-range
of concentrations), we may suspect that model misspecification is not only
due to the varying durations but also caused by over-dispersion. Therefore,
another approach would be to fit a negative-binomial model. Such a model
may be specified using type = "negbin2"; there is also the option "negbin1"
for a slightly different model (Delignette-Muller et al., 2014b).
chlordan.LL.3.nb.we <-drm(repro ~ conc,
data = chlordan,
fct = LL.3(),
type = "negbin2",
weight = time)
Again, we look at the summary output, which should be interpreted in exactly

the same way as for the weighted Poisson model:
summary(chlordan.LL.3.nb.we)
##
##
##
## b:(Intercept) 0.91769 0.11430 8.0291 9.349e-16 ***
## d:(Intercept) 5.57532 0.27033 20.6242 < 2.2e-16 ***
## e:(Intercept) 1.68993 0.27040 6.2498 4.111e-10 ***
## O:(Intercept) -2.18109 0.28539 -7.6426 2.130e-14 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
confint(chlordan.LL.3.nb.we)
## 2.5 % 97.5 %
## b:(Intercept) 0.6936773 1.141707
## d:(Intercept) 5.0454867 6.105160
## e:(Intercept) 1.1599602 2.219908
## O:(Intercept) -2.7404380 -1.621742
Slightly different parameter estimates are obtained, but the standard errors
are almost twice as large as compared to the Poisson model. The param-
eter, O, appearing in the summary output is exp(ω) in Equation (A.8) in
Section A.2.2.2.
One last alternative approach would be to adjust for model misspecifi-
cation in terms of the assumed distribution (while still assuming that the
dose-response model function is correctly specified) by replacing the default
naı̈ve standard errors by robust standard errors (see Section A.5).
coeftest(chlordan.LL.3, vcov. = sandwich)
##
##
## b:(Intercept) 1.08950 0.17078 6.3795 3.413e-08 ***
## d:(Intercept) 113.66634 4.25420 26.7186 < 2.2e-16 ***
## e:(Intercept) 1.61556 0.25454 6.3470 3.862e-08 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
This approach results in exactly the same parameter estimates as for the
unweighted Poisson model but with modified standard errors, which are almost
twice as large. Indeed, the results are similar to the results from the fitted
negative binomial model.
In terms of point estimates and 95% confidence intervals, the above results
for the models taking the weights into account are very similar to the results
originally reported by Delignette-Muller et al. (2014a) but obtained using a
much more complex Bayesian approach.

In this section, we consider one data example with two dose-response curves
and an external control group. An entire dose-response curve corresponds
to a single treatment, but as there are no replicated dose-response curves,

systematic treatment differences cannot be separated from random variation
between dose-response curves. We can still compare treatment groups, as you
will see below, but it will take additional arguments (independent of the dose-
response data) and information to rule out confounding.
3.2.1 Counting bacteria colonies: Wadley’s problem

Trajstman (1989) describes a laboratory experiment on isolation of My-
cobacterium bovis (M. bovis) colonies, which cause bovine tuberculosis. In
short, M. bovis is detected from bovine tissues grown on culture plates if
M. bovis colonies appear after some time. As M. bovis colonies are slow
growing, contaminating bacteria may overgrow the cultures. Decontaminants
may be used for killing contaminating bacteria. However, the dose is impor-
tant to avoid that M. bovis are also killed while ensuring uncontaminated
plates. In this experiment the performance of two decontaminants, oxalic
acid and 1-hexadecylpyridium chloride (HPC), was compared. The dataset
decontaminants is shown below.
head(decontaminants)
## conc count group

## 1 0.75 2 hpc
## 2 0.75 4 hpc
## 3 0.75 8 hpc
## 4 0.75 9 hpc
## 5 0.75 10 hpc
## 6 0.75 1 hpc
summary(decontaminants)
## conc count group

## Min. :0.0000 Min. : 0.00 hpc :69
## 1st Qu.:0.0050 1st Qu.:18.00 oxalic:59
## Median :0.0750 Median :33.00
## Mean :0.5503 Mean :32.27
## 3rd Qu.:0.3750 3rd Qu.:46.00
## Max. :5.0000 Max. :80.00
Ten plates were used at each level of decontaminant concentration. Separately,

20 plates without any decontaminant were used for the control experiment.
The numbers of colonies on the plates were counted.
We revisit the analysis of the data that was carried out by Morgan and
Smith (1992). However, instead of separate models per decontaminant, we fit a
joint model including both decontaminants as well as the control group, which
was common to the two groups. Specifically, we use the argument pmodels
to indicate that the parameter d is shared among the three groups (∼1),
whereas the parameters b and d should be different between the two groups
(∼group-1). Note that this model specification is only possible because the
data have been arranged in such a way that the control group has been merged
with one of the two other groups (it does not matter which one). Following the
original analysis we fit a three-parameter Weibull type 1 model. This model
fitting is an example of Wadley’s problem (Finney, 1971, Chapter 10).
decon.W1.3.po <- drm(count ~ conc,
curveid = group,
data = decontaminants,
fct = W1.3(),
type = "Poisson",
pmodels = list(~ group-1, ~ 1, ~ group-1))
decon.summary <- decontaminants %>%

group_by(group, conc) %>%
summarise(mean.count = mean(count),
sd2.count=var(count))
plot(sd2.count ~ mean.count, data = decon.summary,

ylim = c(0, 160),
xlab = "Average count",
ylab = "Squared standard deviation")
abline(a = 0, b = 1, lty = 2)
The plot of average counts versus squared empirical standard deviations in

Figure 3.7 indicates that the variation increases more with increasing average
counts than can be explained by the Poisson distribution (points tend to be
above the 45◦ reference line), i.e., over-dispersion is present.
Therefore, we expect that the standard errors of parameter estimates re-
ported in the summary output below are too small as the Poisson model is not
able to incorporate all variation in the counts:
summary(decon.W1.3.po)
##
## Model fitted: Weibull (type 1) with lower limit at 0 (3 parms)
##
##
## b:grouphpc 0.773346 0.051621 14.9813 < 2.2e-16 ***
## b:groupoxalic 0.371221 0.039024 9.5127 < 2.2e-16 ***
Squared standard deviation
150
100
50
0
10 20 30 40 50
Average count
FIGURE 3.7
Average counts versus squared empirical standard deviations per concentra-
tion per decontaminant for the decontaminants dataset. The broken line
corresponds to the variation predicted by the Poisson distribution.
## d:(Intercept) 49.848286 1.218586 40.9066 < 2.2e-16 ***

## e:grouphpc 0.268028 0.017093 15.6803 < 2.2e-16 ***
## e:groupoxalic 1.498828 0.238817 6.2761 3.473e-10 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
One way to take over-dispersion into account is by using robust standard

errors.
coeftest(decon.W1.3.po, vcov = sandwich)
##
##
## b:grouphpc 0.773346 0.107492 7.1945 5.387e-11 ***
## b:groupoxalic 0.371221 0.090580 4.0983 7.498e-05 ***
## d:(Intercept) 49.848286 2.506653 19.8864 < 2.2e-16 ***
## e:grouphpc 0.268028 0.039689 6.7531 5.102e-10 ***
## e:groupoxalic 1.498828 0.508578 2.9471 0.003839 **
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
The robust standard errors are approximately 2–3 times larger than the naı̈ve
standard errors shown in the above summary output.
An alternative approach would be to retain the three-parameter Weibull
type 1 model but replace the Poisson distribution by a negative binomial
distribution:
decon.W1.3.nb <- drm(count ~ conc,

curveid = group,
data = decontaminants,
fct = W1.3(),
type = "negbin2",
pmodels = list(~ group-1, ~ 1, ~ group-1))
The resulting summary output looks like this:
summary(decon.W1.3.nb)
##
## Model fitted: Weibull (type 1) with lower limit at 0 (3 parms)
##
##
## b:grouphpc 0.777719 0.083595 9.3034 < 2.2e-16 ***
## b:groupoxalic 0.368470 0.061458 5.9955 2.029e-09 ***
## d:(Intercept) 50.014470 1.988664 25.1498 < 2.2e-16 ***
## e:grouphpc 0.264038 0.026979 9.7868 < 2.2e-16 ***
## e:groupoxalic 1.463080 0.379882 3.8514 0.0001174 ***
## O:(Intercept) 0.505370 0.208863 2.4196 0.0155366 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
We see the same picture: standard errors increased as compared to the Poisson
model, but less dramatically this time: only around a factor 2. The fitted dose-
response curves based on the negative binomial model are shown in Figure 3.8.
The potency of two decontaminants may be compared in terms of the ratio
of their EC50 values using the function EDcomp():
EDcomp(decon.W1.3.nb, c(50, 50), interval = "delta")
##
##
## hpc/oxalic:50/50 0.30459 0.11155 0.49763
plot(decon.W1.3.nb, broken = TRUE, type = "all",

xlim = c(0, 10),
xlab = "Concentration (% w/v)",
ylab = "Colonies")
80
hpc
oxalic
60
Colonies
40
20
0 0.1 10
Concentration (% w/v)
FIGURE 3.8
Three-parameter Weibull type 1 model fitted to the dataset decontaminants.
There is a significant difference in the potency of the two decontaminants (p <

0.0001). Moreover, the decontaminant HPC is approximately 0.305 (0.112,
0.498) times as potent as oxalic acid.
4
Multinomial dose-response data
A multinomial response is usually ordinal, providing a more detailed picture

of gradual change than the one offered by a binary response. A prominent
example is various severity scores ranging from unaffected/alive through one
or more stages of being partially affected to 100% affected/dead.
There exists very little work on the statistical analysis of multinomial dose-
response data. Some authors have used general modeling approaches for multi-
nomial data (Finney, 1971, pp. 220-226; Morgan, 1992, pp. 119–12), whereas
others have proposed novel approaches specifically for dose-response data (Fox
and Landis, 2016).
We will take a different, pragmatic and practical approach towards analysis
of multinomial dose-response data, exploiting that such data may be analyzed
through a series of models fitted to suitably derived binomial dose-response
data. Specifically, if there are K categories in the multinomial dose-response
data, then K − 1 models for binomial dose-response data need to be fitted
(there may be several ways of doing this); the examples will demonstrate how
to do this in practice. Admittedly, this approach is an approximation: The
loss in efficiency from not having a joint model for the multinomial data, but
repeatedly fitting models to derived binomial data, however, seems to be small
in most cases (Begg and Gray, 1984).
One key advantage of fitting multiple models to binomial dose-response
data is that it becomes easier to interpret the results. Another advantage is
that over-dispersion, which can also occur for multinomial data, may be dealt
with the same way as for binomial data. One disadvantage may possibly be
that the obtained results are less parsimonious, but models for multinomial
data also yield results that are more challenging to interpret.
Below we consider two textbook examples with trichotomous dose-response
data, but the approach also works for multinomial data with more categories.
In this chapter we use the following extension packages:
library(drc)
library(devtools)
install_github("doseResponse/drcData")
library(drcData)
install_github("SigneMJensen/mmmVcov")
85
library(mmmVcov)
library(multcomp)
library(dplyr)
4.1 Trichotomous data

4.1.1 Insecticide residues
Finney (1971, pp. 223–226) analyzed data from an experiment investigating
the effect of residues found in alfalfa plants that were sprayed with one of two
preparations (Standard vs. Test) of the insecticide guthion in varying doses
(µg (Standard) and ml (Test)). For each dose, 50 insects were exposed to the
residue for 17 hours. Then the number of alive, moribund, and dead insects
were counted.
guthion
## trt dose alive moribund dead total

## 1 S 20.0 44 1 5 50
## 2 S 35.0 28 1 21 50
## 3 S 45.0 8 7 35 50
## 4 T 1.0 37 1 12 50
## 5 T 1.5 20 2 28 50
## 6 T 2.0 8 6 36 50
We fit the two-parameter log-logistic model to binomial data obtained from the
multinomial data by merging the two categories “alive” and “moribund” into
a single category while retaining the category “dead.” The following analysis
compares “dead” to “moribund” and “alive” combined:
guthion.LL.2.am <- drm(dead/total ~ dose,

curveid = trt,
weights = total,
data = guthion,
fct = LL.2(),
type = "binomial")
Estimated LC50 values and their ratio are obtained as in the previous chapters
using the functions ED() and EDcomp()
Multinomial dose-response data 87
ED(guthion.LL.2.am, c(50))
##
##
## e:S:50 36.891558 1.887823
## e:T:50 1.432554 0.083677
EDcomp(guthion.LL.2.am, c(50, 50), interval = "delta")
##
##
## S/T:50/50 25.752 21.833 29.672
The estimated relative potency is close to the one obtained by Finney (1971,
p. 226) (the difference most likely due to different estimation procedures).
Next, we fit the two-parameter log-logistic model to binomial data obtained
by merging the neighbouring categories “moribund” and “dead” into a single
category while retaining the category “alive” unchanged. In general, you may
want to merge more and more categories starting from 100% affected/dead
to compare decreasing accumulated severity to alive. Or you start from un-
affected/alive and include more and more severe stages step by step. The
following analysis compares “moribund” and “dead” combined to “alive”:
guthion.LL.2.dm <- drm((moribund+dead)/total ~ dose,
curveid = trt,
weights = total,
data = guthion,
fct = LL.2(),
type = "binomial")
Estimated LC50 values and their ratio are obtained as before:

ED(guthion.LL.2.dm, c(50))
##
##
## e:S:50 33.908787 1.552187
## e:T:50 1.324984 0.066308
EDcomp(guthion.LL.2.dm, c(50, 50), interval = "delta")
##
##
## S/T:50/50 25.592 22.190 28.994
Again, the estimated relative potency is close to the one obtained by Finney
(1971, p. 226). Figure 4.1 also shows that the two model fits are quite similar
in terms of differences between the two groups, reflecting that the “moribund”
category contains only little information.
plot(guthion.LL.2.am, broken = TRUE, type = "all",

xlim = c(0, 1000),
ylim = c(0, 1),
xlab = "Dose",
ylab = "Proportion affected")
plot(guthion.LL.2.dm, add = TRUE, col = "gray",

xlim = c(0, 1000),
legend = FALSE)
1.0
S
Proportion affected
0.8 T
0.6
0.4
0.2
0.0
0 1 10 100 1000
Dose
FIGURE 4.1
Two-parameter log-logistic models fitted to binomial data obtained from the
dataset guthion. Black lines for the analysis of “dead” vs. “alive” and “mori-
bund” combined, grey lines for the analysis of “dead” and “moribund” com-
bined vs. “alive.”
Moreover, the fitted binomial dose-response models have fairly similar esti-
mated slope parameters in both model fits (keeping in mind the narrow range
of the doses applied: 1, 45):
coef(summary(guthion.LL.2.am))[1:2, ]

## b:S -3.768548 0.7123614 -5.290220 1.221692e-07
## b:T -3.051117 0.6568528 -4.645054 3.399877e-06
coef(summary(guthion.LL.2.dm))[1:2, ]

## b:S -4.365964 0.7367935 -5.925628 3.111063e-09
## b:T -3.850136 0.7031664 -5.475427 4.364577e-08
(using square brackets to choose the first two rows containing the estimated
slopes). This indicates that the proportional odds assumption would be ap-
propriate for these multinomial data (e.g., Morgan, 1992, p. 120). Note that if
we had instead fitted Weibull type 2 models (using W2.2()) we could have as-
sessed in the same way if a proportional hazards assumption would have been
appropriate. However, in contrast to the analysis reported by Finney (1971,
pp. 223–226), the approach based on analyzing several binomial datasets does
not rely on any proportional odds assumption; we prefer this approach. If a
single pooled estimate of the relative potency is of interest and possibly war-
ranted (as in this case) then it may be derived by averaging the estimates
from the binomial model fits as shown below.
We use the function mjust() from the GitHub package mmmVcov to es-
timate the variance-covariance matrix of the two relative potencies, i.e., to
obtain estimated standard errors and, more importantly, the estimated corre-
lation between the two estimated relative potencies (which are estimated from
two marginal model fits, not from a joint model fit). This is possible using R
but, to our knowledge, not in any other statistical software (Jensen and Ritz,
2015).
relPot.pooled <- mjust(list(guthion.LL.2.am, guthion.LL.2.dm),

list("e_S/e_T", "e_S/e_T"), seType = "san")
relPot.pooled[["covar"]]
## [,1] [,2]
## [1,] 0.4024401 0.5548167
## [2,] 0.5548167 1.8861332
confint(glht(parm(relPot.pooled[["coef"]][,1],
relPot.pooled[["covar"]]),
linfct = matrix(c(0.5, 0.5), 1, 2)))
##
## Simultaneous Confidence Intervals
##
## Fit: NULL
##
## Quantile = 1.96
## 95% family-wise confidence level
##
##
## Estimate lwr upr
## 1 == 0 25.6721 23.8656 27.4786
The argument seType = "san" implies that robust standard errors and cor-
relations are estimated (Jensen and Ritz, 2018). Once the variance-covariance
matrix has been estimated (and stored under the name relPot.pooled)
the functions glht() and parm() from the package multcomp may be used
to calculate the average of the two estimates (specified using linfct =
matrix(c(0.5, 0.5), 1, 2)) and the corresponding estimated standard er-
ror. Finally, confint() produces the associated confidence interval (95% by
default). Note that the confidence interval is narrower than the two confidence
intervals derived from the binomial dose-response models.
4.1.2 Effect of two arboviruses on chicken embryos

Morgan (1992) discusses a trichotomous multinomial data example where
chicken embryos were exposed to increasing doses of an arbovirus
(PFA/embryo) in order to investigate the potency of the virus. The exper-
iment was carried out for two different types of arbovirus (below denoted
“FP” and “T,” respectively). Initially there were 20 embryos per dose, but a
few embryos died of other reasons in the first few days. After 18 days each em-
bryo was assessed and categorised as not deformed, deformed, or dead (three
categories). The entire dataset is shown below.
arbovirus
## dose total dead def trt
## 1 3 17 3 1 FP
## 2 18 19 4 1 FP
## 3 30 19 8 2 FP
## 4 90 20 17 1 FP
## 5 3 19 1 0 T
## 6 20 19 2 0 T
## 7 2400 15 4 9 T
## 8 88000 19 9 10 T
## 9 0 18 1 0 T
We follow the same steps as in the previous subsection: First we fit a two-
parameter log-logistic model to binomial data obtained by merging the not
deformed and deformed categories into a single category:
arbovirus.LL.2.dead <- drm(dead/total ~ dose,

curveid = trt,
weights = total,
data = arbovirus,
fct = LL.2(),
type = "binomial")
Next we fit the model where binomial data were obtained by merging deformed
and dead categories:
arbovirus.LL.2.alive <- drm((def+dead)/total ~ dose,

curveid = trt,
weights = total,
data = arbovirus,
fct = LL.2(),
type = "binomial")
Figure 4.2 shows the fitted dose-response curves with the data. It is apparent
that the estimated slope coefficients are different between the two analyses,
i.e., the proportional odds assumption seems not appropriate. We proceed to
summarize the model fits one by one. First the results for the analysis of dead
vs. deformed+alive:
EDcomp(arbovirus.LL.2.dead, c(50, 50), interval = "delta",

reverse = TRUE)
##
##
## T/FP:50/50 1284.5 -1007.2 3576.1
We use the argument reverse = TRUE to reverse the calculation to have the
reciprocal relative potency estimated (showing how much more potent ar-
bovirus FP is than arbovirus T); by default it would be the other way around.
The estimated relative potency of arbovirus FP relative to arbovirus T is 1284.
In other words the arbovirus FR is approx. 1300 times more potent than the
arbovirus T in killing chicken embryos.
For the second analysis where “deformed” and “dead” were merged we get
the following results:
plot(arbovirus.LL.2.dead, broken = TRUE,

xlim = c(0, 1000000),
ylim = c(0, 1),
xlab = "Dose (plaque-forming assay per embryo)",
ylab = "Proportion affected",
legendPos = c(7, 1))
plot(arbovirus.LL.2.alive, add = TRUE, col = "gray",

legend = FALSE)
1.0
FP
Proportion affected
0.8 T
0.6
0.4
0.2
0.0
0 10 1000 1e+05
Dose (plaque−forming assay per embryo)
FIGURE 4.2
Two-parameter log-logistic models fitted to binomial data derived from the
dataset arbovirus. Black lines for the analysis of dead vs. alive+deformed,
grey lines for the analysis of dead+deformed vs. alive.
EDcomp(arbovirus.LL.2.alive, c(50, 50), interval = "delta",

reverse = TRUE)
##
##
## T/FP:50/50 9.4238 -2.7513 21.5988
The estimated relative potency of arbovirus T relative to arbovirus FP is

9.424. The arbovirus FP is approximately 10 times more potent than the
arbovirus T in being harmful to chicken embryos.
Due to the small sample size in general and the dose range of arbovirus
T not capturing the full response range, the 95% confidence intervals of the
relative potencies have lower limits below 0, which is not meaningful as relative
potencies are by definition positive. These lower limits should be truncated
at 0, indicating that there is only little information in the dose-response data
for estimating relative potencies.
5
Time-to-event-response data
Time-to-event-response data, which are a special type of time-to-event data,

occur frequently in biology, ecology, seed science, and toxicology as the result
of planned experiments.
A time to event (also called an event time) is the time it takes until some
well-defined event happens (e.g., a plant flowering or a seed germinating).
However, the event of interest is usually not observed exactly at the time point
where it took place. This feature usually reflects limitations in the practical
management of experiments and it is only possible to determine timing of
events to have taken place in a certain time interval. Our way of collecting
or observing data leads to a grouping of the time-to-event data according to
the monitoring interval (although the event itself can happen at any time,
in continuous time). Moreover, the event of interest need not occur at all
during the time course of the experiment. This feature is commonly referred
to as right censoring: the experiment had been terminated before the event
would possibly have happened or the event was rendered impossible during
the experiment. Morgan (1992) already pointed out these characteristics but
it seems not to have had much influence on the data analysis in practice,
possibly because of lack of suitable statistical software.
For instance, in germination experiments, time to seed germination is one
outcome of interest. Seeds may only be inspected once a day or week and once
a seed has germinated, the only information obtained is that the event (seed
germination) took place somewhere between the last and the present inspec-
tion. Furthermore, some seeds may remain dormant for the entire duration of
the experiment or some seeds may have been eaten or rotted away during the
experiment.
General time-to-event data are often analyzed by means of semi- or non-
parametric survival analysis methods. In contrast, we will only use fully para-
metric models to analyze time-to-event-response, as they are equally flexible
and more easily provide relevant and interpretable parameter estimates. There
has been a long tradition for analysis of germination data by means of logis-
tic and nonlinear regression techniques although these approaches have some
shortcomings (Morgan, 1992; Ritz et al., 2013b). In particular, these methods
may lead to too small estimated standard errors for parameter estimates of in-
terest (e.g., t50 ) as these techniques do not account for the loss of information
due to grouping and right censoring.
95
Models that are more appropriate need to incorporate time-to-event data,

which are both grouped and right-censored. However, only recently such mod-
els have been developed and, indeed, implemented in R in a general frame-
work, which may be useful for other types of time-to-event-response data, not
only germination data (Ritz et al., 2013b).
The key insight linking the present chapter with the previous chapters on
modeling of continuous, binomial, count, and trichotomous dose-response data
is that cumulative distribution functions are special cases of dose-response
model functions with a few restrictions imposed (the time elapsed since the
beginning of the experiment is “the dose”). In practice, it means we have
nearly the same flexibility for modeling time-to-event-response data as for
any other type of dose-response data.
For convenience, we assume that the experiment begins at time 0. We
formulate the statistical model in terms of the cumulative distribution function
F , which for each time point t ≥ 0 describes the fraction of events that
took place between time 0 and time t. By definition F is 0 at time 0 and
F (t) approaches a fraction d ≤ 1 as more and more time elapses. If d <
1 then we will assume that F has a jump and reaches 1 at infinity, that
is F (∞) = 1 (otherwise F is not a cumulative distribution function). This
modeling assumption implies that there is possibly a proportion of events
(equal to 1 − d) that will not be observed to happen during the experiment
(regardless of the duration of the experiment). In each application, it has to
be judged if this is a reasonable assumption.
We consider model parameterizations of the cumulative distribution func-
tion that ensures both F (0) = 0 and convenient biological interpretations
(Ritz et al., 2013b). This parallels the dose-response modeling approach from
the previous chapters. A commonly used model is the three-parameter log-
logistic model:
d d
F (t) = = b , (5.1)
1 + exp[b{log(t) − log(t50 )}]

t
1+ t50
which is identical to the three-parameter log-logistic function introduced pre-

viously, but now as a function of time (t) instead of dose. The interpretation of
the slope parameter b is unaltered. The parameter t50 is the time point where
50% of the events have happened, that is, by definition, it is relative to the
maximum proportion d observed during the experiment. Thus, the parameter
t50 has the same interpretation in terms of a halfway reduction relative to the
achievable maximum as has the ED50 for continuous data. In particular, this
means that t50 only corresponds to the median event time in case d = 1, in
which case F (t50 ) = 0.5. In general, this will not hold for the model defined by
Equation (5.1). In the general case, t50 corresponds to the 100d/2 percentile
in the distribution of event times. See Section A.2.3 for more details.
In germination experiments, t50 may be interpreted as the median ger-
mination time for the population of viable or non-dormant seeds only. This
Time-to-event-response data 97
interpretation, however, depends on the assumption that all viable or non-

dormant seeds have germinated by the end of the experiment, which may be
established empirically by running the experiment until germination ceases
and the maximum proportion becomes clearly discernible from the data.
library(drc)
library(devtools)
library(drcData)
library(lmtest)
library(metafor)
library(multcomp)
library(plyr)
library(sandwich)
5.1 Analysis of a single germination curve

5.1.1 Germination of Stellaria media seeds
We consider data from a germination test of chickweed (Stellaria media) seeds
(Ritz et al., 2013b). Seeds from a chlorsulfuron-resistant biotype were germi-
nated in petri dishes (with a diameter of 9 cm) in a dark growth cabinet;
we refer to Ritz et al. (2013b) for more details. The information about the
petri dishes is no longer available. Two hundred seeds were placed on filter
plates. At 34 consecutive inspection times, the number of germinated seeds
was recorded and these seeds were removed. Definition of a germinated seed
was the breakthrough of the seed testa by the radicle.
Data are available in the built-in dataset chickweed where there is one
row per monitoring interval (in total 35 intervals when the interval after ter-
mination of the experiment is included). Note that the intervals need not have
the same width and, furthermore, intervals without counts do not contribute
to parameter estimation and could be left out. The first and last three lines
of the dataset are shown below.
head(chickweed, 3)
## start end count

## 1 0 12 0
## 2 12 22 0
## 3 22 30 0
tail(chickweed, 3)
## start end count

## 33 266.5 276.5 1
## 34 276.5 281.5 0
## 35 281.5 Inf 160
The columns named start and end contain the lower and upper limits of the
monitoring intervals, whereas the column named count contains the number
of seeds that germinated in the corresponding monitoring interval. The last
row in the dataset contains the number of seeds that did not germinate during
the experiment, with a monitoring interval ending at infinity (Inf).
Stellaria media has dormant seeds and therefore we would not expect 100%
germination. Consequently, it is relevant to fit a three-parameter log-logistic
model where the upper limit (the parameter d) corresponds to the total pro-
portion germinated. The model is fitted using drm(), but the specification is
slightly different from the one used in the previous chapters. The first argu-
ment in drm() is the model formula linking the response variable (count) on
the left-hand side to the monitoring intervals, which are encoded by means of
two variables (start and end) specified using + on the right-hand side of the
formula. In addition, the argument type is equal to "event" to ensure that
estimation is carried out by assuming the appropriate multinomial model as
briefly explained above.
chickweed.LL.3 <- drm(count ~ start + end,
data = chickweed,
fct = LL.3(),
type = "event")
Before looking at the summary output, we make a residual plot to examine

if the chosen model provides a reasonable description of the trend in the
data. The residual plot in Figure 5.1 shows a clear curvilinear trend where
residuals are first positive, then negative, then positive, and, finally, becoming
negative again. This means the chosen model function is not providing a very
good description of the trend in the data. It seems that there is much more
nonlinearity in the data than captured by an s-shape. However, the residuals
are small, in most cases being smaller than 0.01 (1% percentage point) and the
detected departure will hardly be visible when we plot the fitted germination
curve below, and it may not affect parameter estimates and standard errors
much. It is also possible to construct a QQ plot as seen for continuous data
plot(fitted(chickweed.LL.3), residuals(chickweed.LL.3))
abline(h=0, lty=2)
qqnorm(residuals(chickweed.LL.3))
0.010
residuals(chickweed.LL.3)
0.000
−0.010
0.00 0.05 0.10 0.15 0.20
fitted(chickweed.LL.3)
FIGURE 5.1
Residual plot for the three-parameter log-logistic model fitted to the dataset
chickweed.
but it should be kept in mind that the residuals will only be approximately
normally distributed and the many intervals without germination (if included)
will blur the picture somewhat (see Figure 5.2).
summary(chickweed.LL.3)
##
##
##
## b:(Intercept) -20.76732 2.94421 -7.0536 1.743e-12 ***
## d:(Intercept) 0.20011 0.02830 7.0711 1.537e-12 ***
## e:(Intercept) 196.05291 2.50572 78.2422 < 2.2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Normal Q−Q Plot
0.010
Sample Quantiles
0.000
−0.010
−2 −1 0 1 2
Theoretical Quantiles
FIGURE 5.2
QQ plot of raw residuals from the three-parameter log-logistic model fitted to
the dataset chickweed.
The summary output reveals that only 20 (2.8) percent of the seeds germinated
during the experiment. Moreover, it took 196 (2.5) hours to germinate half of
the seeds that germinated during the experiment.
One way to address this slight model misspecification caused by the cluster-
ing due to use of several petri dishes (even without having information about
which seeds belonged to which dish) is to use robust standard errors (see Sec-
tion A.5 for more explanation). We need to activate the packages lmtest and
sandwich to be able to use the function coeftest() with the argument vcov.
= sandwich.
coeftest(chickweed.LL.3,
vcov. = sandwich)
##
##

## b:(Intercept) -20.76732 7.84206 -2.6482 0.01246 *
## d:(Intercept) 0.20011 0.17438 1.1475 0.25967
## e:(Intercept) 196.05291 4.40685 44.4882 < 2e-16 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
The resulting standard errors increased by a factor between 1.5 and 2.5, re-
flecting that the model was slightly misspecified.
Likewise, estimation of multiple time points corresponding to arbitrary
percentages of germination with robust standard errors may be achieved using
ED() with the same specification as in coeftest(). Below we estimate t10 ,
t50 , and t90 with robust standard errors and 95% confidence intervals.
ED(chickweed.LL.3,
c(10, 50, 90),
interval = "delta",
vcov. = sandwich)
##
##
## e:1:10 176.3697 8.2987 160.1045 192.6349
## e:1:50 196.0529 4.4069 187.4156 204.6902
## e:1:90 217.9328 9.7192 198.8834 236.9821
It is also possible to estimate absolute germination times that refer to a total of

100% germination. As the estimated upper limit was just above 20% (0.20),
it makes sense to estimate t5 , t10 , and t15 , the time it takes to reach 5%,
10%, and 15% germination of all seeds used. Again, we can use ED() but
specifying proportions rather than percentages and using the argument type
= "absolute" to switch from the default calculation relative to the estimated
upper limit(s) to the absolute calculation. Which version is more relevant will
depend on the application, but it should ideally be identified already during
the planning phase of the experiment.
ED(chickweed.LL.3,
c(0.05, 0.1, 0.15),
interval = "delta",
type = "absolute",
vcov. = sandwich)
##
##
## e:1:0.05 185.9445 5.7647 174.6459 197.2431
## e:1:0.1 196.0425 4.4069 187.4053 204.6798
## e:1:0.15 206.6817 6.0136 194.8953 218.4680
Using the method plot, we show the fitted regression curve together with
the cumulated proportions of germinated seeds. This plot may also serve as
a kind of graphical model diagnostics. We add a confidence band based on
95% pointwise Wald-type confidence intervals, which may easily be obtained
using the predict() method by providing a dataset of time points for which
to obtain predicted values (in this case 0:300, corresponding to 0, 1, 2, . . .,
300 hours). Figure 5.3 shows the resulting plot. The fitted germination curve
seems to describe the increase in germination over time adequately.
As Wald-type confidence intervals are used for constructing the confidence
band, the lower limit of the band would become slightly negative for around
180 hours if it were not truncated at 0. However, by default the predict()
method has the argument constrain switched on so that truncation is en-
forced (whenever it is reasonable).
plot(chickweed.LL.3,
xlim = c(0, 300), ylim = c(0, 0.30),
xlab = "Time (hours)",
ylab = "Proportion germinated",
log = "")
chickweed.predicted <- predict(chickweed.LL.3,

data.frame(0:300),
interval = "confidence")
lines(0:300, chickweed.predicted[, "Lower"],

col = "darkgray")
lines(0:300, chickweed.predicted[, "Upper"],

col = "darkgray")
5.2 Analysis of data from multiple germination curves

It may often be relevant to compare germination among different groups or
treatments. The corresponding experimental design will often lead to data for
multiple germination curves, ideally per group or treatment. Had the response
been continuous, a nonlinear mixed-effects dose-response model could possibly
0.30
Proportion germinated 0.25
0.20
0.15
0.10
0.05
0.00
0 100 200
Time (hours)
FIGURE 5.3
The germination curve based on the three-parameter log-logistic model fit-
ted to the dataset chickweed plotted together with the observed cumulated
proportions and a pointwise 95% confidence band.
have been fitted. To our knowledge no equivalent parametric random-effects

models have been proposed for time-to-event data. A very flexible alternative
is the two-step approach proposed by Jensen et al. (2017), extending the ap-
proach proposed by Jiang and Kopp-Schneider (2014) used in Subsection 1.2.2.
We briefly outline the approach:
The first step consists of fitting a separate time-to-event-response model
to data from each independent sub-experiment (e.g, batch, pot, or tray). From
each model fit, estimates of parameters of interest (with corresponding esti-
mated standard errors) are then collected in a new dataset. Information on
all other variables related to the experimental design should also be gathered
in the same dataset.
In the second step these estimates are used as response values for a meta
analysis, which may be carried out by fitting a linear mixed model. In contrast
to a usual linear mixed model, the residual standard error, which reflects
variation within a sub-experiment, is not estimated. Instead, it is assumed
to be known and equal to the estimated standard errors corresponding to
the estimates. Inferential procedures such as pairwise comparisons based on
approximate Wald-type u-tests are available, just like for an ordinary linear
mixed model.
The approach is applied in two examples. In the first example, time-to-

event-response curves were obtained for a number of doses without replication
of germination. So it is a one-way layout without replication at the treatment
level and, consequently, treatment is confounded with the between-curve vari-
ation. The second example is a large germination experiment with a three-way
layout including replicated germination sub-experiments for each combination
of the three treatments.
5.2.1 Time to death of daphnias

In a toxicity test the effect of cadmium chloride (Cd) was examined by ex-
posing young (<24 hours) daphnias (Daphnia magna) to different doses of
Cd (Kooijman, 1981). Daphnias were held in hard water at 20◦ C for 3 weeks
with the water and Cd refreshed on day 2, 4, 7, 9, 11, 14, 16, 18 and 21.
Seven different doses were applied: 0.0, 3.2, 5.6 10.0, 18.0, 32.0, and 56.0 µg
Cd L−1 water. For each dose, approximately 50 daphnias were exposed. When
refreshing the resolution, the number of surviving daphnias were counted. The
dataset is shown below.
head(CadmiumDaphnia)
## Dose Time Total Start End Dead

## 1 0.0 2 50 0 2 0
## 2 0.0 4 50 2 4 0
## 3 0.0 7 50 4 7 0
## 4 0.0 9 50 7 9 0
## 5 0.0 11 50 9 11 0
## 6 0.0 14 50 11 14 0
The dataset contains four variables: Dose denotes the dose of cadmium chlo-
ride, Start and End denote the limits on the monitoring intervals, and, finally,
Dead denotes the number of dead daphnias recorded for the corresponding
monitoring interval. Now we proceed to analyze these data using the meta-
analytic approach.
5.2.1.1 Step 1
We will assume that a two-parameter log-logistic model will fit data for all
doses. This may be highly questionable as all doses are not equally toxic.
However, the duration of the experiment was not long enough to observe
such differences. Moreover, it is not a limitation: different models for differ-
ent doses is also possible as long as the same parameter of interest may be
estimated.
We start out by manually fitting a two-parameter log-logistic model sepa-
rately to data for each dose, and then estimated parameters from the resulting
model fits are combined. It is a somewhat repetitive task, but it sheds some
light on how the meta-analytic approach works. So we start out fitting 7 mod-
els, one per dose, using the function subset() for defining anonymous subsets
on the fly.
daphnia.LL.2.00.0 <- drm(Dead ~ Start + End,
data = subset(CadmiumDaphnia, Dose == "0.0"),
fct = LL.2(),
type = "event")

fct = LL.2(),
type = "event")

fct = LL.2(),
type = "event")

fct = LL.2(),
type = "event")

fct = LL.2(),
type = "event")

fct = LL.2(),
type = "event")

fct = LL.2(),
type = "event")
The next step is to extract the estimated t50 and the corresponding standard
error from each of the above model fits. For the two-parameter log-logistic
model it is the estimate for the parameter e. We store the results in a dataset
called cadmium.t50, which we initially define as an empty dataset.
cadmium.t50 <- data.frame(t50 = rep(NA, 7),
t50.se = rep(NA, 7))
Then the dataset is filled in row by row using summary() in combination with
coef() for extracting estimates and standard errors.
cadmium.t50[1, ] <- coef(summary(daphnia.LL.2.00.0))[2, 1:2]
The manual fitting of 7 models could have been avoided using a call of drm()
with the argument curveid to assume different model parameters for different
doses. The R lines are shown below but are not executed; we invite the reader
to do so.
daphnia.LL.2 <- drm(Dead ~ Start + End,

curveid = as.factor(Dose),
data = CadmiumDaphnia,
fct = LL.2(),
type = "event")
cadmium.t50x <- coef(summary(daphnia.LL.2))[8:14, 1:2]

colnames(cadmium.t50x) <- c("t50", "t50.se")
In theory, exactly the same parameter estimates and standard errors could
then be obtained in one go. However, in practice, parameter estimates and
standard errors may be slightly different as fitting this model involves estima-
tion of 7 + 7 = 14 parameter simultaneously. In general, it is more challenging
to fit such simultaneous models as lack of convergence and convergence to sub-
optimal parameter estimates may often occur, i.e., the estimation procedure
may have problems in case many parameters need to be estimated. Assuming
a shared parameter across doses (a shared slope parameter b) would be an
option (by means of the argument pmodels used in the same way as previ-
ously in Subsection 1.2.1). However, we believe it will in practice be difficult
to justify such assumptions based on a priori knowledge about the underly-
ing biological mechanisms. Moreover, as you will see in the next example (in
Subsection 5.2.2), fitting separate models to subsets of a dataset may be au-
tomated since it better scales up for more complex experimental designs, also
when it comes to merging relevant information about the experimental design
with the estimates obtained. To ensure that all relevant information is carried
over to the second step, we add a column to the dataset with information on
the doses.
cadmium.t50 <- data.frame(Dose = c(0, 3.2, 5.6, 10, 18, 32, 56),
cadmium.t50)
cadmium.t50
## Dose t50 t50.se

## 1 0.0 23.412632 6.3833146
## 2 3.2 74.466687 69.9816272
## 3 5.6 38.331515 10.3035074
## 4 10.0 35.720487 8.2251577
## 5 18.0 11.034994 1.0383054
## 6 32.0 5.192387 0.3324163
## 7 56.0 1.954142 0.1575264
The scatterplot in Figure 5.4 shows the estimated t50 values as a function of
dose. A distinct nonlinear trend is discernible but the variation in the estimates
time-to-event seems to depend on the magnitude of the estimates: the larger
the estimate, the larger the variation.
5.2.1.2 Step 2
We use the package metafor to combine estimates from the individual model
fits into pooled estimates through the meta-analytic approach (Viechtbauer,
2010). Specifically we use the function rma(), which takes the estimated
t50 as the first argument and the corresponding squared standard errors as
the second argument. The explanatory variables are specified through a for-
mula in the argument mods. Finally, the argument data is used for provid-
ing the dataset where the variables given in the first three arguments are
found. Specifically, we fit a weighted version of a linear regression model for
the logarithm-transformed t50 as a function of dose through the specification
mods = Dose, exploiting that the estimated standard error for a logarithm-
transformed estimate is the estimated standard error of the untransformed
variable divided by the estimate itself (which may be obtained using the delta
method). The reason for applying a logarithm transformation is that the sum-
mary output showing increased variation with increased estimated t50 values
and, at the same time, this transformation may remedy some of the observed
nonlinearity seen in the below plot (Figure 5.4).
cadmium.t50.log.oneway <- rma(log(t50),

(t50.se/t50)^2,
mods= ~ Dose,
data = cadmium.t50)
plot(t50 ~ Dose,
data = cadmium.t50,
xlab = "Dose (mu g Cd/L)",
ylab = "Time to 50% died")
60
Time to 50% died
40
20
0
0 10 20 30 40 50
Dose (mu g Cd/L)
FIGURE 5.4
Estimated t50 values from the first step in the meta-analytic approach as a
function of dose of cadmium chloride (using the dataset CadmiumDaphnia).
coef() applied to the summary output provides a condensed version of the

summary output, only containing information on the parameter estimates for
the linear regression.
coef(summary(cadmium.t50.log.oneway))
## estimate se zval pval ci.lb

## intrcpt 3.63528901 0.234205039 15.521822 2.469372e-54 3.17625557
## Dose -0.05508716 0.007800124 -7.062344 1.637174e-12 -0.07037512
## ci.ub
## intrcpt 4.0943225
## Dose -0.0397992
The logarithm-transformed t50 decreases significantly with increasing dose

(p < 0.0001): for each increase in the dose of 1 µg Cd L−1 the average t50 value
is reduced by 5.4 percent (we can claim a relative reduction as consequence
of using a logarithm transformation). It is worth mentioning that the meta-

analytic approach on untransformed estimated t50 values (assuming a linear
trend in dose) showed no significant effect of dose.
The predict() method in metafor can be used to obtain predicted values
for the fitted regression line along with the corresponding (pointwise) confi-
dence intervals. The argument newmods is used for specifying the doses for
which a prediction is needed. We ask for predicted values for doses between 0
and 60 at increments of 1.
cadmium.predicted <- predict(cadmium.t50.log.oneway,

newmods = seq(0, 60, by = 1))
head(cadmium.predicted)
## pred se ci.lb ci.ub cr.lb cr.ub

## 1 3.6353 0.2342 3.1763 4.0943 2.8436 4.4270
## 2 3.5802 0.2283 3.1328 4.0276 2.7952 4.3652
## 3 3.5251 0.2225 3.0891 3.9611 2.7466 4.3037
## 4 3.4700 0.2168 3.0452 3.8949 2.6977 4.2424
## 5 3.4149 0.2112 3.0010 3.8289 2.6485 4.1814
## 6 3.3599 0.2058 2.9565 3.7632 2.5991 4.1206
The predicted values are found in the first column named pred, whereas the
third and fourth columns contain the lower and upper limits of the (95%) con-
fidence intervals (named ci.lb and ci.ub, respectively). Note that predicted
values and confidence intervals are back-transformed from the logarithmic
scale to the original time scale using the exponential function exp().
So we can redraw the scatterplot with the fitted nonlinear curve and con-
fidence band (Figure 5.5). The fitted regression curve describes the average
trend in the data quite satisfactorily.
5.2.2 A hierarchical three-way factorial design

We consider data from a greenhouse trial studying seedling emergence for
an herbicide-susceptible (S) and an herbicide-resistant (R) blackgrass sub-
population, both isolated from a single population (Keshtkar et al., 2017).
For two different temperature regimes, seedling emergence in the two sub-
populations was investigated under four different sowing depths. For each of
the 16 combinations of depth, sub-population, and temperature, four pots
with 36 seeds were used and germination was monitored for 22–43 days at
intervals of varying lengths for the two temperatures (21–22 monitoring inter-
vals). The pots were arranged in a completely randomized design with four
repetitions of each combination of the three factors. This completely random-
ized experiment was conducted twice. Thus, the experimental design was a
completely randomized three-way factorial layout with 2 blocks. We refer to
Keshtkar et al. (2017) for more information on the experiment.
plot(t50 ~ Dose,
data = cadmium.t50,
xlab = "Dose (mu g Cd/L)",
ylab = "Time to 50% died")
lines(0:60,
exp(cadmium.predicted[["pred"]]), lty = 1) # fitted line
lines(0:60,
exp(cadmium.predicted[["ci.lb"]]), lty = 2) # lower CI
lines(0:60,
exp(cadmium.predicted[["ci.ub"]]), lty = 2) # upper CI
Time to 50% died
60
40
20
0
0 10 20 30 40 50
Dose (mu g Cd/L)
FIGURE 5.5
Estimated t50 values from the first step in the meta-analytic approach plotted
vs. doses of cadmium chloride (using the dataset CadmiumDaphnia). The fitted
and back-transformed regression curve (solid line) with a 95% confidence band
(dashed lines) is also shown.
Data are in the dataset blackgrass. The three treatments are encoded in
the variables Bio, Depth, and Temp.
head(blackgrass)
## Exp Temp Popu Bio Depth Rep Start.Day End.Day Ger Accum.Ger TotalSeed
## 1 1 10 914 S 0 1 0 360 0 0 36
## 2 1 10 914 S 0 1 360 376 0 0 36

## 3 1 10 914 S 0 1 376 384 0 0 36
## 4 1 10 914 S 0 1 384 400 0 0 36
## 5 1 10 914 S 0 1 400 408 0 0 36
## 6 1 10 914 S 0 1 408 424 0 0 36
For the first experiment, the data may be tabulated as follows.
with(subset(blackgrass, Exp == 1), table(Depth, Bio, Temp))
## , , Temp = 10
##
## Bio
## Depth R S
## 0 88 88
## 1 88 88
## 3 88 88
## 6 88 88
##
## , , Temp = 17
##
## Bio
## Depth R S
## 0 84 84
## 1 84 84
## 3 84 84
## 6 84 84
The numbers 88 and 84 correspond to 22 or 21 monitoring intervals in four

replicates (pots), respectively.
The experimental design is encoded through the variables Exp and Rep,
where the latter is the identifier variable for pots in the dataset. It is, however,
not unique as multiple instances of Rep equal to 1 occur (in different sub-
experiments) as seen from the following table.
with(blackgrass, table (Exp, Rep))
## Rep
## Exp 1 2 3 4
## 1 344 344 344 344
## 2 344 344 344 344
Therefore, we start out adding such a variable (given the name Pot) to the
dataset blackgrass. It would be nice if such a unique identifier were always
included in such datasets as it may help understand where the random varia-
tion in the data is introduced.
blackgrass[["Pot"]] <- with(blackgrass,

as.numeric(interaction(Exp, Bio, Depth,
Temp, Rep)))
Pot has 128 levels, corresponding to a total of 128 pots being used in the
experiment. Now we are ready to begin analyzing the data using the two-step
approach.
5.2.2.1 Step 1
A three-parameter log-logistic model is fitted to data from each pot sepa-
rately. We show how to do this in an automated way, avoiding fitting a model
manually for each pot.
Looping through data from all pots and fitting a dose-response model to
data from each pot separately may effectively be carried out using the R
package plyr in combination with the following helper function that defines
what has to be done for each pot (in this case fitting an LL.3() model). Note
that if you need to use different model functions within the same experiment
(e.g., both LL.2() and LL.3()) then automation may become more difficult
as a systematic approach for model selection is needed. In the code below
dataSet is a placeholder for the subset provided to the function applied for
each pot.
fitFct.LL.3 <- function(dataSet)

{
modelFit <- try(drm(Ger ~ Start.Day + End.Day,
data = dataSet,
fct = LL.3(),
type = "event"), silent = TRUE)
if (inherits(modelFit, "try-error")) {modelFit <- NULL}
return(modelFit)
}
The model may not converge for all pots, which is the reason why we use try()
inside the helper function; this function will catch all errors and, thereby,
ensure that fitting a lot of models in a loop, as shown below, will not terminate
prematurely due to a few cases with lack of convergence and resulting in no
model fits. For germination data, lack of convergence can happen because of
a low number of seedlings emerging or because all emerging seedlings emerge
at the same or at very few times. In both cases, it may be difficult to fit any
dose-response model. Sometimes manual fitting for problematic cases may still
result in a useful model fit being obtained, but it is more cumbersome and, if
different model functions are assumed, results may be not be fully comparable
between automatic and manual model fits.
We use the function dlply() from plyr for the actual looping: For data cor-
responding to each combination of the three treatment variables Bio, Depth,
and Temp and the grouping variables Exp and Pot, reflecting the hierarchical
design, the above-defined function fitFct.LL.3() is applied. Note that com-
bining these 5 variables yield many more combinations than actually present
in the dataset, but it ensures that the experimental design is carried along with
the estimates. dlply() takes a dataset (data frame) as input and returns a
list.
black.grass.modelfits2 <- dlply(blackgrass,

.(Exp, Bio, Depth, Temp, Pot),
fitFct.LL.3)
The list black.grass.modelfits2 is quite large and you may not want to
look at it as it is. We proceed to extract the three parameter estimates and
corresponding estimated standard errors from each model fit. For that pur-
pose, we use another helper function as defined below.
paramFct.LL.3 <- function(fitObj)

{
if (is.null(fitObj)) {return(rep(NA, 6))}
# handling replicates with all values missing
coefSum <- coef(summary(fitObj))

returnVec <- c(coefSum[1, 1:2], # slope
coefSum[2, 1:2], # maximum
ED(fitObj, 50, display = FALSE)[1:2]) # t50
names(returnVec) <- c("b", "b.se",

"d", "d.se",
"t50", "t50.se")
# adding column names
returnVec
}
Again, we exploit functionality from the extension package plyr for looping
through the list of model fits (black.grass.modelfits2). This time we use
the function ldply(), which takes a list and returns a dataset. Note also that
we estimate t50 values that are relative to the estimated upper limits (this
could be changed using the argument type = "absolute" as seen previously
in Subsection 5.2.1).
blackgrass.parms <- ldply(black.grass.modelfits2, paramFct.LL.3)
summary(blackgrass.parms)
## Exp Bio Depth Temp Pot

## Min. :1.0 R:64 Min. :0.00 Min. :10.0 Min. : 1.00
## 1st Qu.:1.0 S:64 1st Qu.:0.75 1st Qu.:10.0 1st Qu.: 32.75
## Median :1.5 Median :2.00 Median :13.5 Median : 64.50
## Mean :1.5 Mean :2.50 Mean :13.5 Mean : 64.50
## 3rd Qu.:2.0 3rd Qu.:3.75 3rd Qu.:17.0 3rd Qu.: 96.25
## Max. :2.0 Max. :6.00 Max. :17.0 Max. :128.00
##
## b b.se d
## Min. :-321.593 Min. : 0.4799 Min. :-1041.3839
## 1st Qu.: -13.253 1st Qu.: 1.6114 1st Qu.: 0.4171
## Median : -10.042 Median : 2.1744 Median : 0.6113
## Mean : -22.468 Mean : 34.8117 Mean : -7.9552
## 3rd Qu.: -8.183 3rd Qu.: 2.7119 3rd Qu.: 0.7780
## Max. : 2.735 Max. :847.1601 Max. : 1.7379
## NA's :6 NA's :6 NA's :6
## d.se t50 t50.se
## Min. : 0.02737 Min. :201.5 Min. : 1.764
## 1st Qu.: 0.06949 1st Qu.:241.0 1st Qu.: 7.269
## Median : 0.07857 Median :440.7 Median : 17.958
## Mean : 1.52145 Mean :429.8 Mean : 28.087
## 3rd Qu.: 0.08271 3rd Qu.:549.2 3rd Qu.: 26.911

## Max. :173.79139 Max. :823.2 Max. :528.089
## NA's :6 NA's :6 NA's :6
From the summary of the resulting dataset named blackgrass.parms we see

that there are 6 missing values (NAs) in the dataset. Therefore, for some pots
the three-parameter log-logistic model could not be fitted.
5.2.2.2 Step 2
The second step of the analysis is fitting the meta-analytic random-effects
model for the parameter(s) of interest. We only consider a model for t50.
However, the same approach could be applied to the analysis of the slope and
maximum germination parameters.
Specifically, we consider a model that can be formulated as follows.
Estimated t50,i = µ(Temperaturei , Bioi , Depthi ) + A(experimenti ) + θi + εi ,
where i refers to the independent sampling units in the experiment, the µ

values denote the parameters signifying the mean t50 levels for the treatment
combinations of the interaction; A is a random effect capturing differences
between experiments; θ is the random effect explaining the heterogeneity be-

tween pots (nested within experiment); and ε is the residual error. Random
effects and residual errors are assumed to be mutually independent and nor-
mally distributed with mean 0. The standard deviations of the random effects
are unknown and have to be estimated from data and the standard deviations
of the residual errors are assumed to be known and equal to the estimated
standard errors of the estimates. Hence, the smaller the standard error in
a pot-specific analysis in Step 1, the larger the weight of the corresponding
estimate in the meta-analytic model fit in Step 2.
To estimate the average time to 50% emergence for each combination of the
three treatments, a new variable that corresponds to the three-way interaction
of Bio, Depth, and Temp is added to the dataset. Introducing such a variable
will allow us to define relevant pairwise comparisons, which is much more
interesting than testing interactions in the three-way factorial layout.
blackgrass.parms[["BioDepthTemp"]] <-
with(blackgrass.parms, interaction(Bio, Depth, Temp))
We use the function rma.mv() in the package metafor (Viechtbauer, 2010) for
estimating the random-effects meta-analytic model. The default method for
estimation is REML.
blackgrass.t50.mm <- rma.mv(t50, (t50.se)^2,

mods= ~ BioDepthTemp - 1,
random= ~ 1|Exp/Pot,
data = blackgrass.parms)
## Warning in rma.mv(t50, (t50.se)^2, mods = ~BioDepthTemp - 1,

random = ~1 | : Rows with NAs omitted from model fitting.
The specification Exp/Pot ensures that the hierarchical structure of the ex-
perimental design is also incorporated in the statistical analysis. A warning
message is issued because missing values occurred and were left out of the
analysis. It reflects that for some pots the three-parameter log-logistic model
could not be fitted. However, as pointed out previously, it may still be possible
to fit models for these pots manually.
To obtain a condensed summary output with estimates, standard errors,
and confidence intervals, the coef() method may be used.
coef(summary(blackgrass.t50.mm))[, c("estimate", "se",

"ci.lb", "ci.ub")]
## estimate se ci.lb ci.ub

## BioDepthTempR.0.10 684.3207 11.769466 661.2529 707.3884
## BioDepthTempS.0.10 656.2772 11.674943 633.3947 679.1596
## BioDepthTempR.1.10 503.2574 12.036954 479.6654 526.8494

## BioDepthTempS.1.10 478.0808 10.285295 457.9220 498.2396
## BioDepthTempR.3.10 521.1635 10.938573 499.7243 542.6027
## BioDepthTempS.3.10 464.3360 9.918119 444.8968 483.7751
## BioDepthTempR.6.10 605.4620 21.880176 562.5776 648.3463
## BioDepthTempS.6.10 579.8046 11.720794 556.8322 602.7769
## BioDepthTempR.0.17 374.9668 11.346542 352.7279 397.2056
## BioDepthTempS.0.17 375.9767 10.723558 354.9590 396.9945
## BioDepthTempR.1.17 227.7426 8.527922 211.0281 244.4570
## BioDepthTempS.1.17 213.1403 8.024740 197.4121 228.8685
## BioDepthTempR.3.17 227.5065 8.058943 211.7112 243.3017
## BioDepthTempS.3.17 211.6396 7.969696 196.0193 227.2599
## BioDepthTempR.6.17 296.5767 11.850070 273.3510 319.8024
## BioDepthTempS.6.17 261.9117 9.510475 243.2716 280.5519
The summary output (see below) also contains the estimated standard de-
viations for the experiment and pot-specific random effects. These estimates
would be useful to report in a publication to give some insight in how ran-
dom variation is split into between-experiment and between-pot variation (we
observe that the standard deviation describing the between-pot variation is
approximately a factor 3 larger than the standard deviation describing the
between-experiment variation).
summary(blackgrass.t50.mm)[["sigma2"]]
## [1] 44.10246 305.80208
All pairwise comparisons may conveniently be defined using the following

contrast matrix:
allPairWiseComp <- contrMat(coef(blackgrass.t50.mm), "Tukey")
All pairwise comparisons with unadjusted p-values are then calculated (but
not shown) using glht().
blackgrass.allpairwise <- glht(blackgrass.t50.mm,

linfct = allPairWiseComp)
To have all comparisons shown, the summary() method may be used.
summary(blackgrass.allpairwise,
test = adjusted(type = "none"))
We do not show the output as it would contain numerous pairwise compar-

isons, of which only a few may be relevant. Note that multiplicity adjust-
ment of p-values was switched off manually (using test = adjusted(type =
"none") as adjustment is applied by default). It is much better to ask for cer-

tain relevant comparisons. Among other things, Keshtkar et al. (2017) looked
at comparisons between sub-populations for each depth and each temperature.
To obtain the relevant comparisons we define a targeted contrast matrix.
targetedPairWiseComp <-
c("BioDepthTempR.0.10 - BioDepthTempS.0.10 = 0",
"BioDepthTempR.1.10 - BioDepthTempS.1.10 = 0",
"BioDepthTempR.6.17 - BioDepthTempS.6.17 = 0")
The names of combinations as found in the summary output of the model fit
should be used. It is also important to include -1 in the model specification
as otherwise the parameterization will not allow the above specification of
contrasts. We can use glht() and the corresponding summary() method to
show the comparisons.
blackgrass.targeted.pairwise <- glht(blackgrass.t50.mm,
linfct = targetedPairWiseComp)
summary(blackgrass.targeted.pairwise)
##
## Simultaneous Tests for General Linear Hypotheses
##
## Fit: rma.mv(yi = t50, V = (t50.se)^2, mods = ~BioDepthTemp - 1,
## random = ~1 | Exp/Pot, data = blackgrass.parms)
##
## Estimate Std. Error z value
## BioDepthTempR.0.10 - BioDepthTempS.0.10 == 0 28.043 15.186 1.847
## BioDepthTempR.0.17 - BioDepthTempS.0.17 == 0 -1.010 14.119 -0.072
## Pr(>|z|)
## BioDepthTempR.0.10 - BioDepthTempS.0.10 == 0 0.414905
## BioDepthTempR.3.10 - BioDepthTempS.3.10 == 0 0.000131 ***

## BioDepthTempR.6.17 - BioDepthTempS.6.17 == 0 0.087475 .
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
## (Adjusted p values reported -- single-step method)
We only find a significant difference between resistant and susceptible sub-

populations for depth 3 cm and the 10/5◦ C temperature regime: in the resis-
tant sub-population seedlings emerged 57 (13) hours later.
6
Benchmark dose estimation
Benchmark doses are in a sense parallel to the concept of effective doses (see
Section A.13) (Ritz et al., 2013a), but while the latter usually occurs in a
dose-region with a reasonable amount of data available, the benchmark dose
estimation is essentially an interpolation method for the low-dose region where
little or no data is present.
To define a benchmark dose (BMD) we first need to define the background
level, p0 , as the probability of an adverse response for an unexposed popula-
tion. This level could be taken from the literature, estimated from data, or
pre-specified at a fixed level such as 0.05. A benchmark response or benchmark
risk (BMR) is a small increase above the background level in the probability
of an adverse event. The BMD is the dose eliciting a response equal to BMR
on average, assuming some dose-response model.
Usually risk assessment is not based on the BMD itself but on the bench-
mark dose lower limit (BMDL), which is defined as the lower limit of the
confidence interval for the BMD (Crump, 1984). As the estimate of BMDL is
partly determined by the uncertainty of the BMD estimate, it penalizes poor
designs and analysis strategies.
The BMDL may be found using different approaches. The simplest ap-
proach proposed by Crump (Crump, 1984) is using a one-sided Wald-type
confidence interval. However, it may result in negative BMDLs. One simple
solution may be to combine the Wald-type confidence intervals with a trans-
formation (typically the logarithm) to avoid negative values (Buckley et al.,
2009; Namata et al., 2008; Moon et al., 2013). Other approaches for estimat-
ing the BMDL include finding the dose associated with the upper limit of the
confidence band of the fitted dose-response curve (inverse regression) (Buckley
et al., 2009; Fang et al., 2015), different bootstrap strategies (Buckley et al.,
2009; Piegorsch et al., 2012, 2014; Zhu et al., 2007) and profile likelihood in-
tervals (Yu and Catalano, 2005; Izadi et al., 2012; Fox et al., 2017; Ringblom
et al., 2014).
library(drc)
library(devtools)
library(drcData)
119
install_github("DoseResponse/bmd")
library(bmd)
library(metafor)
library(sandwich)
6.1 Binomial dose-response data

For all binomial responses, BMD may be defined in the same way as data
share a common response scale (Crump, 1984). Two common definitions are
used: Added or additional risk:
BM R = f (BM D, β) − p0 (6.1)
Excess or extra risk:

f (BM D, β) − p0
BM R = (6.2)
1 − p0
The excess risk definition is recommended by several authorities (U. S. EPA,
2012; Hardy et al., 2017). When the background risk is 0 the two definitions
are identical and thus result in the same BMD.
6.1.1 Pathogens in food

A total of 108 healthy human volunteers participated in a study of the effect of
different doses of the pathogen, Echovirus 12 . Each participant was exposed
to the pathogen in one of four doses. Based on measurement of blood samples,
it was determined whether or not the exposure had led to an infection. Data
were reported by Moon et al. (2013). The full dataset is shown below. The
dose of the pathogen is reported in plague forming units (pfu).
head(echovirus)
## dose total infected

## 1 330 50 15
## 2 1000 20 9
## 3 3300 26 19
## 4 10000 12 12
As seen from the data, more people were exposed to the lower doses than
to the higher. This has two major advantages: Fewer people will experience
Benchmark dose estimation 121
an adverse effect and we get more information about the low dose area, which
we are interested in when considering benchmark doses.
The purpose of this study was to estimate the BMD10 , i.e., to find the
safe dose associated with a predefined acceptable level, BMR = 10% increase
over the background level, of pathogen-specific infections. For the dataset
echovirus, this involves extrapolation, which we would usually recommend to
avoid. However, in the present case one could argue that we also know that
no one will be infected unless they are exposed to the specific pathogen. This
means we have the extra information that a potential dose 0 would result in
no infections. In practice, this information is incorporated in our model fitting
by assuming the background level is 0. Accordingly, we fit a two-parameter
log-logistic model.
pathogen.m1 <- drm(infected/total ~ dose,

weights = total,
data = echovirus,
fct = LL.2(),
type = "binomial")
In this case, we are interested in finding the dose that results in 10% being
infected. Figure 6.1 shows the data together with the estimated dose-response
model (R lines for the plot are provided in Subsection C.2.1). The figure
illustrates how BMD can be found as the dose resulting in the level of risk
determined by the pre-specified level of BMR.
We use the package bmd to estimate the BMD and the corresponding
BMDL as follows:
bmd(pathogen.m1,
0.10,
def = "additional",
backgType = "modelBased")
## BMD BMDL
## 90.32084 10.0272
The estimated BMD resulting in 10% being infected is 90.3 pfu with a BMDL
of 10.0 pfu based on a Wald-type confidence interval that relies on asymptotic,
large-sample results on the behaviour of parameter estimates.
The default method for finding BMDL in bmd() is using the lower limit
of Wald-type confidence intervals. An alternative approach for estimating
BMDL is to use a bootstrap method, which does not, to the same degree,
depend on asymptotic results. We choose a parametric bootstrap approach
where new datasets are generated from a binomial distribution with parame-
ters (Ni , Ni /Yi ) for each dose, i, separately. Here Ni refers to the total number
in dose group i and Yi refers to the number of infected in dose group i. The
1.0
0.8
Proportion infected
0.6
0.4
0.2
0.0
BMD 1000 10000
Echovirus 12 (pfu)
FIGURE 6.1
Two-parameter log-logistic model fitted to the dataset echovirus. With a back-
ground level of 0, BMD is the dose associated with the risk equal to the
pre-specified BMR of 0.10.
BMDL is defined as the 5% percentile in the bootstrap distribution of BMDs

(using the percentile interval).
pathogen.bmd.boot <- bmdBoot(pathogen.m1,

0.10,
def = "additional",
backgType = "modelBased",
bootType = "parametric",
R = 1000,
display = FALSE)
pathogen.bmd.boot$Result
## BMD BMDL
## 90.32084 28.78989
The BMDL found by the bootstrap method is much higher than the BMDL
we found above. It may be caused by the relatively few doses in the present
study and the fact that we work with binomial data. A simple histogram of
bootstrap estimates (Figure 6.2) shows that the distribution is skewed and
140
100
Frequency
60
0 20
0 100 200 300 400
Bootstrap estimates
FIGURE 6.2
Histogram of the distribution of the bootstrap estimates used for estimating
BMDL for the dataset echovirus.
further investigations will reveal a dependence between the estimate and the
bootstrap standard error. This phenomenon is well known when working with
distributions like the binomial where the variance is a function of the mean
(Hesterberg, 2015). Consequently, using the percentile bootstrap confidence
interval can give misleading estimates of the BMDL.
As an alternative to the default percentile interval, the function bmdBoot()
has an option called “BCa” for the argument bootInterval using a bias-
corrected and adjusted bootstrap interval (DiCiccio and Efron, 1996).
bmdBoot(pathogen.m1,
0.10,
def = "additional",
bootType = "parametric",
R = 1000,
bootInterval = "BCa")
## BMD BMDL
## 90.32084 27.51616
The corrected bootstrap interval results in almost the same BMDL, indi-
cating that the method based on the adjusted bootstrap interval may not be
needed here.
6.1.2 Chromosomal damage

Bentley et al. (2000) report findings from a genetic study that examined the
potential chromosomal damage (as non-disjunction events) in human lym-
phocytes after in vitro exposure of cells to carbendazim, a potentially potent
metabolite of the agricultural fungicide benomyl. It is relatively easy to work
with a large number of healthy lymphocytes and consequently, sample sizes
can grow quickly in such studies. For instance, Bentley et al. (2000) worked
with 2000 cells for each of 13 dose levels.
We will assume a background level of chromosomal damage different from
0, above 0. Additionally, we do not expect all cells to be damaged. Therefore,
we fit a four-parameter log-logistic model such that both lower and upper
limits will be estimated.
carbendazim.m1 <- drm(damage/total ~ dose,
weights = total,
data = carbendazim,
fct = LL.4(),
type = "binomial")
There are only data in the low dose region, i.e., the region of doses leading
to small effects (Figure 6.3). Therefore it is possible to look at small BMRs. For
this example, we will consider BMR = 0.01. The background risk is estimated
by the model to be 0.00825. Assuming additional risk, the BMD is the dose
resulting in a proportion of damaged cells equal to 0.00825 + 0.01 = 0.01825.
bmd(carbendazim.m1,
0.01,
def = "additional",
## BMD BMDL
## 756.442 683.1871
Alternatively, we could consider the excess risk definition. Still using the
estimated background level, BMD is the dose associated with a proportion of
cells being damaged of 0.01816 (0.01·(1-0.00825)+0.00825).
bmd(carbendazim.m1,
0.01,
def = "excess",
## BMD BMDL
## 754.9196 682.057
plot(carbendazim.m1,
broken = TRUE,
xlab = "Carbendazim (ng/ml)",
ylab = "Proportion of damaged cells")
Proportion of damaged cells
0.05
0.04
0.03
0.02
0.01
0 100 1000
Carbendazim (ng/ml)
FIGURE 6.3
Four-parameter log-logistic model fitted to the dataset carbendazim.
For this example with a low background risk, the difference between
additional and excess is small. Looking at the definition of BMD in (6.1)
and (6.2) this is not surprising. For p0 ≈ 0 they both reduce to BM R =
f (BM D, β).
Finally, if we knew from the literature that the background level is actually
0.01, then the BMD associated with a BMR = 0.01 is found by specifying a
user-defined background level, which will overrule any estimated background
level. The BMD is then the dose associated with a proportion of cells being
damaged equal to 0.0199 (0.01·(1-0.01)+0.01).
bmd(carbendazim.m1,
0.01,
def = "excess",
backgType = "absolute",
backg = 0.01)
## BMD BMDL
## 785.8377 703.7944
6.1.3 Tumor incidence continued: Integration of historical

data
We return to the data from Subsection 2.1.2 describing liver tumor incidence
in female Sprague-Dawley rats that were exposed to a chemical like 2,3,7,8-
tetrachlorodibenzo-p-dioxin (TCDD). A two-parameter log-logistic model was
fitted to data.
liver.tumor.LL.2 <- drm(incidence/total ~ conc,

weights = total,
data = liver.tumor,
fct = LL.2(),
type = "binomial")
Finding the BMD corresponding to a BMR = 0.1 using excess risk is then
straightforward.
BMD.NTP <- bmd(liver.tumor.LL.2,

0.1,
def = "excess",
display = FALSE)
BMD.NTP$Results
## BMD BMDL
## 23.90412 20.651
The data are from a National Toxicology Program (NTP) study (National
Toxicology Program, 2006). Other studies, however, examined the same end-
points using the same species, sex, and exposure method. For instance, the
dataset below is from a similar study examining the effect of TCDD on liver
tumors in female Sprague-Dawley rats (Kociba et al., 1978).
head(TCDD)
## conc total incidence

## 1 0.00 86 2
## 2 1.55 50 1
## 3 7.15 50 9
## 4 38.56 45 14
liver.tumor.Kociba.LL.2 <- drm(incidence/total ~ conc,

weights = total,
data = TCDD,
fct = LL.2(),
type = "binomial")
BMD.Kociba <- bmd(liver.tumor.Kociba.LL.2,

0.1,
def = "excess",
display = FALSE)
BMD.Kociba$Results
## BMD BMDL
## 4.795546 1.175081
Combining estimates from dose-response models can be done using a meta-

analytic approach: We construct a new dataset collecting estimates and stan-
dard errors for BMD from all (here only two) studies. Standard errors are
calculated from BMDL knowing these were based on Wald-type confidence
intervals.
step2Data <- data.frame(BMD = c(BMD.NTP$Results[1],

BMD.Kociba$Results[1]),
SE = c(BMD.NTP$SE,
BMD.Kociba$SE))
For the meta-analysis approach, we use the package metafor. In meta-

analysis, estimation involves weights, which depend on the standard errors as
well as the between-study variation. If we want to put more weight on the
new dataset, we can instead provide our own weights. For our current data
example, we choose to assign less weight to older studies.
step2Data$my.weights <- c(1,0.5)

meta.m1 <- rma(BMD,

SE^2,
data = step2Data,
weights = my.weights,
level = 0.9)
meta.m1
##
## Random-Effects Model (k = 2; tau^2 estimator: REML)
##
## tau^2 (estimated amount of total heterogeneity): 178.1907
## (SE = 258.1913)
## tau (square root of estimated tau^2 value): 13.3488
## I^2 (total heterogeneity / total variability): 97.60%
## H^2 (total variability / sampling variability): 41.70
##
## Test for Heterogeneity:
## Q(df = 1) = 41.7000, p-val < .0001
##
## Model Results:
##
## estimate se zval pval ci.lb ci.ub
## 17.5346 10.0634 1.7424 0.0814 0.9818 34.0874 .
##
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
By combing the two studies we find that the estimated BMD is equal to
17.53 ng/kg and the corresponding estimated BMDL is equal to 0.98 ng/kg.
6.2 Continuous dose-response data

For continuous responses, things are more complicated as there is no longer
a natural probability scale for the response between 0 and 1. One possi-
ble straightforward definition of BMD is to base it upon a relative calcu-
lation (Murrell et al., 1998):
f (BM D, β) − f (0, β)
BM R = (6.3)
f (∞, β) − f (0, β)
This definition resembles that of effective doses for continuous dose-response
data. However, the definition does not incorporate any background level due
to the occurrence of adverse events.
The so-called hybrid approach leads to more sensible definitions incorpo-

rating a background level (Gaylor and Slikker, 1990; Kodell and West, 1993;
Budtz-Jørgensen et al., 2001). Basically, the hybrid approach defines the BMD
as the dose corresponding to a predefined increase in risk/probability of falling
below (or exceeding) a certain cut-off value on the response scale. The cut-
off, x0 , dividing values of the continuous response into ranges of normal and
adverse values may be defined in several ways. It may be based on expert
knowledge or using a certain percentile in the distribution of the unexposed
population.
Assuming a normal distribution for the unexposed population, the back-
ground risk, p0 , which is the probability of experiencing an adverse event for
a member of the unexposed population, can be calculated as follows:

x0 − f (0, β)
p0 = 1 − Φ (6.4)
σ
for a dose-response model function f , described by parameters β and Φ, which

denotes the cumulative distribution function of a standard normal distribu-
tion. Similarly, we consider the probability of an adverse response in an ex-
posed population. Based on the hybrid approach, the BMD, defined in terms
of additional risk, is given as the solution to the following equation:

x0 − f (BM D, β)
BM R = 1 − Φ − p0 (6.5)
σ
Likewise, for excess risk, BMD is the solution to the equation:

1 − Φ x0 −f (BM
σ
D,β)
− p0
BM R = (6.6)
1 − p0
With these definitions in place we consider two examples.
6.2.1 Toxicity of copper in an ecosystem with giant kelp

In an ecotoxicological study of heavy metal contamination, Giant kelp, Mac-
tocystis pyrifera, was exposed to different concentrations of copper. Eight dif-
ferent concentrations were used with up to 5 replicates per concentration. The
response was the length of the germination tubes, the shorter the tube the
more toxic was the heavy metal contamination. Data are taken from Chapman
et al. (1995).
head(GiantKelp)
## tubeLength dose
## 1 19.58 0.0
## 2 18.75 0.0
## 3 19.14 0.0
## 4 16.50 0.0
## 5 17.93 0.0
## 6 18.26 5.6
A four-parameter log-logistic model is fitted to data.
kelp.m1 <- drm(tubeLength ~ dose,

data = GiantKelp,
fct = LL.4())
As expected we see a decreasing trend in the length of the germination

tubes with an increasing concentration of copper in the water environment
(Figure 6.4).
plot(kelp.m1,
type = "all",
broken = TRUE,
xlab = expression(paste("Copper ", mu, "g/L",sep="")),
ylab = "Length germination tube (mm)")
Length germination tube (mm)
20
18
16
14
12
10
8
6
0 10 100
Copper µg/L
FIGURE 6.4
Four-parameter log-logistic model fitted to the dataset GiantKelp.
To estimate the BMD and corresponding BMDL, we first need to define

what an adverse event is. Such a definition could be based on expert knowl-
edge. For instance, a length of germination tube shorter than 14 mm could
imply inability to function properly. Then we would use 14 mm as the cut-off.
The BMD and corresponding BMDL, associated with a BMR = 0.1, can now
be found using the function bmd(). We specify the background risk based on
an absolute level of the response, which is length of the germination tubes.
This means that we specify the background using the argument backgType =
"absolute".
bmd(kelp.m1,
0.1,
backgType = "absolute",
backg = 14,
def = "hybridAdd")
## BMD BMDL
## 12.30724 5.248608
We could also have used the cut-off for dichotomizing the continuous re-
sponse variable. Then everything would seemingly become simpler in the sense
that we could stick to the binomial dose-response model where BMD is per-
haps more conveniently defined.
GiantKelp$tubeLengthBin <- with(GiantKelp,

ifelse(tubeLength < 14, 1, 0))
kelp.m2 <- drm(tubeLengthBin ~ dose,

data = GiantKelp,
fct = LL.4(),
type = "binomial")
bmd(kelp.m2,
0.1,
def = "additional")
## BMD BMDL
## 9.278657 -53.68048
The estimated BMD becomes somewhat smaller, but the estimated BMDL
becomes much smaller, taking on a negative value. So, by dichotomizing the
data, we lose much information. In this example so much information is lost
that we in principle are not able to say anything about the BMD; it could be
any value.
In practice, a cut-off for defining an adverse event is rarely known. Instead

we can use 2 standard deviations (SD) based on the background distribution
to define an adverse event. The argument backType = "hybridSD" specifies
that the background levels should be based on the SD for the control group.
The default is to use 2·SD.
bmd(kelp.m1,
0.1,
backgType = "hybridSD",
def = "hybridAdd")
## BMD BMDL
## 8.878006 2.542278
Finally, looking at the residual plot in Figure 6.5 we may doubt the validity
of the assumption of variance homogeneity.
plot(resid(kelp.m1) ~ predict(kelp.m1),
ylab = "Residuals",
xlab = "Predicted")
abline(h = 0)
4
2
Residuals
0
−2
8 10 12 14 16 18
Predicted
FIGURE 6.5
Residual plot for the four-parameter log-logistic model fitted to the dataset
GiantKelp.
A simple way to address the apparent model misspecification is to base

the BMDL on robust standard errors (see Section A.5 for more details).
bmd(kelp.m1,
0.1,
def = "hybridAdd",
sandwich.vcov = TRUE)
## BMD BMDL
## 8.878006 1.275386
In this case the robust standard errors are larger and the estimated BMDL
becomes smaller.
6.2.2 Toxicity of an antituberculosis drug

Aconiazide is a drug against tuberculosis. Beland et al. (1995) reported data
from a study examining the systemic toxicity of the drug in male Fisher 344
rats. The response was the weight change over a 6-month period with one of
four daily doses received by gavage.
A four-parameter log-logistic model fits the data adequately as seen in
Figure 6.6.
aconiazide.m1 <- drm(weightChange ~ dose,
data = aconiazide,
fct = LL.4())
The summary output is shown below.
summary(aconiazide.m1)
##
##
##
## b:(Intercept) 1.11216 0.45229 2.4589 0.01737 *
## c:(Intercept) -46.62436 821.98122 -0.0567 0.95499
## d:(Intercept) 360.39895 4.96987 72.5168 < 2e-16 ***
## e:(Intercept) 1106.79578 3076.78714 0.3597 0.72054
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
##
##
plot(aconiazide.m1,
broken = TRUE,
type = "all",
xlab = "Aconiazide (mg/kg)",
ylab = "Weight change (g)")
Weight change (g)
350
300
250
0 100
Aconiazide (mg/kg)
FIGURE 6.6
Four-parameter log-logistic model fitted to the dataset aconiazide.
We use again the hybrid approach to estimate the BMD and BMDL for a
BMR = 0.05 but this time the background risk is based on a 3 SD cut-off. As an
alternative to the Wald-type confidence intervals used in the previous example,
we use an inverse regression approach (see Section A.9 and Section A.10). The
function bmd() is specified as follows.
bmd(aconiazide.m1,
0.05,
backg = 3,
def = "hybridAdd",
interval = "inv")
## BMD BMDL
## 97.61672 68.73968
While the four-parameter log-logistic model provided a good fit to data, it

could be argued that a three-parameter model with a lower limit fixed at 0 is
biologically justifiable because extremely high doses of aconiazide may cause
the rat to die immediately, resulting in a weight change of 0 g. We also note
that the estimated parameter c in the above model fit was negative, indicating
that there is no information in the data to estimate this parameter. Below we
fit this reduced model and evaluate how much the estimated BMD and BMDL
are changed.
aconiazide.m2 <- drm(weightChange ~ dose,

data = aconiazide,
fct = LL.3())
bmd(aconiazide.m2,
0.05,
backg = 3,
def = "hybridAdd",
interval = "inv")
## BMD BMDL
## 97.24473 75.20937
The hybrid approach based on the three-parameter log-logistic model fit

results in an estimated BMD of similar magnitude as before but with a higher
estimated BMDL, reflecting that fewer parameters were fitted. This example
reinforces that the choice of model function is crucial as it may affect the
BMDL.
6.3 Model averaging

Interpolation in a dose region with little or no data is highly dependent on
the fit of a dose-response model. In studies estimating BMD and BMDL the
choice of dose-response model describing data can be crucial for the conclu-
sions (Budtz-Jørgensen et al., 2001; Sand et al., 2002). To increase robustness
of the benchmark dose approach, it has been proposed to evaluate several
models and subsequently select the best model determined by means of some
goodness-of-fit criterion (Slob, 2002). However, this procedure does not incor-
porate the uncertainty pertaining to the model selection process and results
based on a “best fitting” model found by a model selection procedure may
result in biased estimates of BMD and too high (non-protecting) estimates
of BMDL (West et al., 2012; Ringblom et al., 2014). As a consequence, it

has been repeatedly advocated to use model averaging for estimation of BMD
and BMDL (Kang et al., 2000; Bailer et al., 2005; Wheeler and Bailer, 2007;
Faes et al., 2007; Namata et al., 2008; Wheeler et al., 2008; Jensen and Ritz,
2015).
Model averaging means forming a model-averaged estimated BMD as a
weighted average of the estimates of BMD obtained from all separately fit-
ted candidate models. Estimation of BMD through model averaging implies
that flexibility of candidate models to some extent compensates for lack of
data.
P Specifically, for K candidate models and assuming weights wk such that
k wk = 1, the formula of the weighted average is shown below.
K
X
BMDMA = wk BMDk
k=1
One choice of weights is the AIC weights defined as follows:
exp(−AICk )
w k = PK
k=1 exp(−AICk )
The above definition of a model-averaged estimated BMD is straightfor-

ward to estimate. In contrast, it is more complicated to estimate the corre-
sponding BMDL; it has been the subject of much methodological research.
Current practice is to use either a type of bootstrap approach (Wheeler and
Bailer, 2007; Wheeler et al., 2008; Buckley et al., 2009; Moon et al., 2013), a
simple average of estimates of BMDL from the candidate models (Kang et al.,
2000), or an approximation based on a variance inequality (Buckland et al.,
1997; Faes et al., 2007; Jensen and Ritz, 2015).
An alternative to averaging estimates of BMD is to make a weighted aver-
age of all candidate model fits, i.e., an average of the estimated dose-response
curves (Wheeler and Bailer, 2007). Below we provide examples of both model
averaging approaches.
6.3.1 Pathogens in food revisited

We return to the Echovirus 12 data from Subsection 6.1.1. Recall that we
previously fitted a two-parameter log-logistic model to data. However, other
models may also provide a reasonable or perhaps even better fit to the data.
Therefore, we fit two additional models to the same data: a two-parameter
log-normal model and a two-parameter Weibull model.
1.0
LL.2
Proportion infected 0.8 LN.2
W2.2
0.6
0.4
0.2
0.0
0 100 1000 10000 1e+05
Echovirus 12 (pfu)
FIGURE 6.7
Three different two-parameter models fitted to the dataset echovirus.
pathogen.LN.2 <- drm(infected/total ~ dose,

weights = total, data = echovirus,
fct = LN.2(), type = "binomial")
pathogen.W2.2 <- drm(infected/total ~ dose,

weights = total, data = echovirus,
fct = W2.2(), type = "binomial")
Figure 6.7 indicates that the Weibull model may be the better choice in
this case (R lines for the plot are provided in Subsection C.2.1).
Besides the visual assessment of the different model fits, AIC may also be
used for comparing the model fits.
pathogen.AIC <- AIC(pathogen.m1, pathogen.LN.2, pathogen.W2.2)

pathogen.AIC
## df AIC
## pathogen.m1 2 18.31155
## pathogen.LN.2 2 17.99485
## pathogen.W2.2 2 16.80155
Based on the AIC values (smaller is better), the Weibull model provides
the better fit to data. As differences between AIC values are small (less than
10), one might suspect that choice of model only has little impact on the
resulting estimated BMD and BMDL values. However, when comparing the
results from the three model fits, large differences are found between estimated
BMD and BMDL values.
bmd.LL.2 <- bmd(pathogen.m1,

0.10,
def = "additional",
display = FALSE)
bmd.LN.2 <- bmd(pathogen.LN.2,
0.10,
def = "additional",
display = FALSE)
bmd.W2.2 <- bmd(pathogen.W2.2,
0.10,
def = "additional",
display = FALSE)
bmd.LL.2$Results
## BMD BMDL
## 90.32084 10.0272
bmd.LN.2$Results
## BMD BMDL
## 102.3873 19.13779
bmd.W2.2$Results
## BMD BMDL
## 60.85477 -5.673325
Notice that the BMDL estimated from the Weibull model is –5.67. A nega-
tive estimated BMDL is not meaningful since a dose cannot be < 0. In practice,
a negative value could be truncated at 0, meaning that we cannot be sure that
any dose above 0 result in an added risk less than 10%. The possibility of get-
ting negative values is a drawback of using Wald-type confidence intervals for
estimating BMDL. An alternative approach is to use inverse regression. Esti-
mating BMD and BMDL using inverse regression results in positive estimates
for all 3 models.
bmd.LL.2.inv <- bmd(pathogen.m1, 0.10, def = "additional",

backgType = "modelBased", interval="inv", display = FALSE)
bmd.LN.2.inv <- bmd(pathogen.LN.2, 0.10, def = "additional",
bmd.W2.2.inv <- bmd(pathogen.W2.2, 0.10, def = "additional",
bmd.LL.2.inv$Results
## BMD BMDL
## 90.32084 40.08553
bmd.LN.2.inv$Results
## BMD BMDL
## 102.3873 49.28374
bmd.W2.2.inv$Results
## BMD BMDL
## 60.85477 22.06095
The Weibull model still results in a much lower estimated BMDL compared
to the other models. In this case, both the choice of dose-response model
function and how to derive the BMDL really matters for the conclusion.
Instead of reporting the BMDL from a single “best fitting,” model as we
did above, we can use model averaging. Choosing AIC weights results in the
following weights, which show that all three models will contribute to the
weighted average.
AICWeights <- exp(-pathogen.AIC$AIC) /

sum(exp(-pathogen.AIC$AIC))
AICWeights
## [1] 0.1449426 0.1989451 0.6561123
Model averaging can be carried out using the function bmdMA(), which as
the first argument, takes a list of the models. The second argument specifies
the weights to be used. These can be directly specified by the user or indirectly
as shown below. Arguments for how BMD and BMR are specified follow the
style of the function bmd(). Finally, we need to specify how to estimate the
BMDL. The choice type = "Kang" results in an estimated BMDL being the
simple weighted average of the estimated BMDL values from the separately
fitted candidate models.
bmdMA(modelList = list(pathogen.m1,pathogen.LN.2,pathogen.W2.2),
modelWeights = "AIC",
bmr = 0.1,
def = "additional",
interval = "inv",
type = "Kang")
## BMD_MA BMDL_MA
## 73.38836 30.08932
6.3.2 Toxicity of an antituberculosis drug revisited

We revisit the example on aconiazide from Subsection 6.2.2. In this experi-
ment, only relatively low doses are provided, leaving room for a lot of vari-
ation in the high-dose area. Consequently, several models may describe the
data adequately. We consider a total of four different three-parameter models;
the log-logistic model already fitted to data and the fit saved in the object
aconiazide.m1, a log-normal model, and two different Weibull models.
aconiazide.LN.3 <- drm(weightChange ~ dose,
data = aconiazide, fct = LN.3())
aconiazide.W1.3 <- drm(weightChange ~ dose,
data = aconiazide, fct = W1.3())
aconiazide.W2.3 <- drm(weightChange ~ dose,
data = aconiazide, fct = W2.3())
The four models provide similar fits to data in the low-dose region
(Figure 6.8, R lines provided in Subsection C.2.2), while deviating more
when extrapolating to high doses as seen in Figure 6.9. A model-averaged
estimated BMD with the corresponding BMDL being estimated using non-
parametric bootstrap may be obtained using bmdMA() with the argument type
= "bootstrap". Non-parametric bootstrap means that new datasets are gen-
erated by resampling with replacement from the original dataset. As before, we
estimate the BMD associated with a BMR = 0.05 using the hybrid approach
with a 3 SD cut-off.
bmdMA(modelList = list(aconiazide.m2, aconiazide.LN.3,
aconiazide.W1.3, aconiazide.W2.3),
bmr = 0.05,
backg = 3,
def = "hybridAdd",
type = "bootstrap")
## BMD_MA BMDL_MA
## 98.46404 70.32581
Weight change (g)

350
300
LL.4
LN.4
W1.4
250 W2.4
0 100
Aconiazide (mg/kg)
FIGURE 6.8
Four different three-parameter models fitted to the dataset aconiazide.
The resulting estimated BMDL is based on the percentile approach, which

is more appropriate for the present example with continuous dose-response
data than for binomial dose-response data as discussed previously in Sub-
section 6.3.1. However, we can repeat the analysis using a bias correction
approach in order to compare results.

bmr=0.05,
back = 3,
def = "hybridAdd",
type = "bootstrap",
bootInterval = "BCa")
## BMD_MA BMDL_MA
## 98.46404 70.32321
400
Weight change (g)
300
200
100
Individual curves
Model−averaged curve
0
0 100 1000 10000
Aconiazide (mg/kg)
FIGURE 6.9
Four different three-parameter models fitted to the dataset aconiazide plot-
ted together with the model-averaged estimated dose-response curve.
The resulting estimated BMDL, 70.32, is close to the estimated BMDL

using the percentile approach, 70.33, indicating that the correction made
through the BCa procedure was small. In theory, the best option is to al-
ways use some kind of adjustment, like BCa, when estimating BMDL using
bootstrap. In practice, however, most adjustments add computation time to
what may already be a time-consuming analysis. This is especially the case
for large datasets.
Finally, we consider model averaging using the approach of averag-
ing the entire fitted dose-response curves. Figure 6.9 shows the model-
averaged estimated dose-response curve together with the four individual es-
timated dose-response curves (R lines for the plot are provided in Subsec-
tion C.2.2).
Estimating BMD and BMDL based on model averaging of all curves can
be achieved using the argument type = "curve".

bmr = 0.05,
back = 3,
def = "hybridAdd",
type = "curve")
## BMD_MA BMDL_MA
## 98.76187 70.72517
It is noteworthy that the change of method not only results in a different

estimated BMDL but also in a different estimated BMD.
7
Hierarchical nonlinear models
Most dose-response experiments are not designed with a simple random

sampling structure for each observation. Hierarchical or multilevel designs
are common, collecting observations repeatedly from the same individual or
grouping individuals into blocks or subunits, e.g., different assays, animal
cages, or laboratories. Any factor that restricts a random sampling design
needs to be included in the dose-response model; therefore, for a hierarchical
design, the correlation between observations from the same experimental unit
needs to be taken into account.
Certain correlation structures may readily be implemented within the
framework of drc, e.g., correlation that is a function of the dose-response
model function (Strodl Andersen et al., 1998). The package medrc extends
many of the capabilities of drc to correlated dose-response data fitted through
nonlinear mixed-effects regression models. Specifically, medrc utilizes the ca-
pabilities of the package nlme (Pinheiro and Bates, 2000); additionally, medrc
offers the same user-friendly interface for estimating and comparing derived
ED50 or benchmark dose parameters.
7.1 Normally distributed dose-response data

A simple hierarchical design consists of observing several response values
within several groups (xij , yij ), with i = 1, . . . ,nj observations within the
group j = 1, . . . , M , or in vector notation xj , y j . Then within each group j
a separate nonlinear model can be assumed:

y j = f j xj , β j + j
The effects between the groups are modeled on the level of the individual
parameters for each group
β j = Aj β̃ + B j bj
where the matrix Aj defines the vector of unknown population parameters β̃

and the indicator matrix B j determines the contribution of each observations
to the group effects bj . Similar to the assumption of normally distributed
145

residuals ij ∼ N 0, σ 2 , we can assume that the group effects also follow a
normal distribution
bj ∼ N (0, Ψ)
summarizing their effect on each parameter as a variance component. With
this assumption of modeling the individual parameters as random effects, the
variance on the scale of each parameter is decomposed into between- and
within-group variability. Further, the parametrization of the individual effects
through variance components allows us to estimate population effects from
individual samples with different content of information, e.g., unbalanced de-
signs with different numbers of observations for each individual or different
dose allocation.
Estimating the unknown parameters in the hierarchical nonlinear model
by maximum likelihood is a challenge. In Appendix A.7 two-stage estimators
for β are discussed.
7.2 The R package medrc

A full maximum likelihood and restricted maximum likelihood implementation
of nonlinear mixed-effects models is available in the R package nlme (Pinheiro
and Bates, 2000). Parameters are estimated using the Lindstrom-Bates algo-
rithm with the function nlme(), where the user has to specify the nonlinear
function and the design matrices for fixed and random effects using a formula
interface.
For simple hierarchical dose-response experiments, the package medrc pro-
vides a user-friendly interface similar to the function drm(), using the package
nlme in the background. All the functionality of the package drc to estimate
and compare effective dose levels is extended to hierarchical models, but re-
stricted to the use of only a single fixed-effects grouping factor.
library(devtools)
install_github("DoseResponse/drc")
install_github("DoseResponse/medrc")
library(medrc)
A more flexible two-stage estimation approach is available with the function

metadrm(), which combines estimates from individual drm() fits using the R
package metafor (Viechtbauer, 2010). As a restriction, the function metadrm()
can include only a single random grouping factor in the second stage.
Hierarchical nonlinear models 147
7.2.1 In vitro effects of the fungicide vinclozolin

Nellemann et al. (2003) carried out experiments to assess the in vitro effects
of the fungicide vinclozolin. Chinese hamster ovary cells were exposed to nine
different concentrations of vinclozolin (in µM) for 22 hours and the resulting
luminescence effects were recorded in an AR reporter gene assay (in lumi-
nescence units). The basic dose-response experiment was repeated 6 times
on different days. The same nine concentrations were used in all six assays.
However, one observation was missing in one assay.
head(vinclozolin)
## exper conc effect

## 1 10509 0.000 1003
## 2 10509 0.025 908
## 3 10509 0.050 997
## 4 10509 0.100 744
## 5 10509 0.200 567
## 6 10509 0.390 314
A plausible biological assumption was that no signal would be observed for
very high fungicide concentrations, indicating that a three-parameter log-
logistic model (Section B.1.1.1) with a lower asymptote at 0 would be a realis-
tic model for the data. The individual effects of each repeated experiment can
be added as normally distributed random effects onto all three parameters:
the slope, asymptote, and ED50.
The model coefficients can be estimated with the two-stage approach, us-
ing the function metadrm(). The interface uses the same arguments as the
drm() function in package drc, but additionally, the second stage grouping
needs to be defined by a factor that is included with the argument ind. The
random effect variance-covariance is defined with the argument struct, using
the default of an unstructured matrix, estimating all random effect variances
and covariances.
metamod <- metadrm(effect ~ conc,
data = vinclozolin,
fct = LL.3(),
ind = exper,
struct = "UN")
summary(metamod)
##
## Two-stage meta-analysis dose-response model
## limit at 0
##
## Call:
## metadrm(formula = effect ~ conc, fct = LL.3(), ind = exper,

## data = vinclozolin, struct = "UN")
##
## Variance estimates:
## estim sqrt
## tau^2.1 0.0316 0.1779
## tau^2.2 687777.9233 829.3238
## tau^2.3 0.0069 0.0830
##
## rho.b:(I rho.d:(I rho.e:(I
## b:(Intercept) 1 -0.8745 0.9315
## d:(Intercept) -0.8745 1 -0.6382
## e:(Intercept) 0.9315 -0.6382 1
##
##
## Coefficients:
## Estimate Std.Err t value Pr(>|t|)
## b:(Intercept) 5.4042e-01 7.8816e-02 6.8568 5.444e-06 ***
## d:(Intercept) 1.9684e+03 3.4107e+02 5.7711 3.692e-05 ***
## e:(Intercept) 1.0312e-01 3.6090e-02 2.8573 0.01199 *
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
The estimates for the population parameters, slope (b), asymptote (d), and
ED50 (e), are presented in the Coefficients section of the summary out-
put together with corresponding standard errors and hypothesis tests for each
coefficient being different to zero. The Variance estimates section contains
the estimated between-assay variance on the scale of each of the three param-
eters, together with the estimated standard deviation as the square root of
the variance.
Functions to obtain the effective dose in the package drc can be directly
applied to the medrc object; here, we are looking at the estimated ED25,
ED50, and ED75 and their corresponding confidence intervals derived by the
delta method.
ED(metamod, respLev = c(25, 50, 75), interval = "delta")
##
##
## e::25 0.0135045 0.0083568 -0.0043075 0.0313165
## e::50 0.1031208 0.0360900 0.0261969 0.1800448
## e::75 0.7874328 0.1531680 0.4609629 1.1139026
The effective dose estimates are derived from the estimated population pa-
rameters and, therefore, these estimates may also be interpreted as population
averages.
As an alternative to the two-stage approach, the full likelihood can be
maximized with the Lindstrom-Bates algorithm using function medrm().
mod <- medrm(effect ~ conc,
data = vinclozolin,
fct = LL.3(),
random = b + d + e ~ 1|exper,
start = c(0.5, 2000, 0.1))
summary(mod)
## Nonlinear mixed-effects model fit by maximum likelihood

## Model: effect ~ meLL.3(conc, b, d, e)
## Data: data
## AIC BIC logLik
## 701.486 721.1889 -340.743
##
## Random effects:
## Formula: list(b ~ 1, d ~ 1, e ~ 1)
## Level: exper
## Structure: General positive-definite, Log-Cholesky
## parametrization
## StdDev Corr
## b 0.15226038 b d
## d 758.87840398 -0.893
## e 0.06970198 0.859 -0.537
## Residual 100.09082654
##
## Fixed effects: b + d + e ~ 1
## Value Std.Error DF t-value p-value
## b 0.5382 0.0706 45 7.624322 0.0000
## d 1987.3111 321.4787 45 6.181782 0.0000
## e 0.1014 0.0322 45 3.144618 0.0029
## Correlation:
## b d
## d -0.819
## e 0.812 -0.517
##
## Standardized Within-Group Residuals:
## Min Q1 Med Q3 Max
## -1.94006282 -0.51813906 -0.09170977 0.49904958 2.20590189
##
## Number of Observations: 53
## Number of Groups: 6
The output is structured in the same way as for the metadrc object with
the population parameter estimates in the Fixed effects section and the
between-assay standard deviation estimates in the Random effects section.
The pairwise correlation between random effects are presented as the lower
triangular part of the correlation matrix directly beside the estimated variance
components.
The function ED() can be directly applied to the mixed-effects model ob-
ject:
ED(mod, respLev = c(25, 50, 75), interval = "delta")
##
##
## e:1:25 0.0131600 0.0071288 -0.0011895 0.0275096
## e:1:50 0.1013562 0.0313061 0.0383402 0.1643722
## e:1:75 0.7806264 0.1410174 0.4967731 1.0644798
We obtain almost similar effective dose-estimates as for the two-stage es-

timation approach; only the standard error estimates are a bit smaller for the
full maximum likelihood method.
7.2.2 Inhibition of photosynthesis in spinach

Streibig et al. (1999) investigated the inhibition of photosynthesis in response
to two synthetic photosystem II inhibitors, the herbicides diuron and benta-
zon. In an experiment, the effect of oxygen consumption of thylakoid mem-
branes (chloroplasts) from spinach was measured after incubation with the
synthetic inhibitors. Five assays, three treated with bentazon and two with
diuron, were used. For each assay, six increasing herbicide concentrations were
applied together with a negative control, using different dose ranges on a log-
arithmic scale for the two treatments to encompass the whole dose-response
range based on preliminary experiments.
spinach$CURVE <- as.factor(spinach$CURVE)

head(spinach)
## CURVE HERBICIDE DOSE SLOPE

## 1 1 bentazon 0.00 1.81295
## 2 1 bentazon 0.00 1.86704
## 3 1 bentazon 0.00 1.95606
## 4 1 bentazon 0.62 1.39073
## 5 1 bentazon 0.62 1.15721
## 6 1 bentazon 0.62 1.06126
For the comparison of the two herbicides, dose-response curves were fitted as-
suming a three-parameter log-logistic model (Section B.1.1.1) with a separate
set of slope, upper asymptote, and ED50 parameters for each of the two treat-
ments. Individual assay effects were included on the slope, upper asymptote,
and ED50 parameters to model the between-assay variability in the different
model scales. Using the information about the between-assay variability is es-
pecially advantageous as the dose levels for the two herbicides did not cover
the same dose range.
The curveid argument in function drm() defines a set of model parame-
ters for every level of a categorical predictor variable, which groups together
observations within an individual. When we want to estimate a separate set
of parameters for each herbicide, we need to extend the indicator matrix Aj
adding indicator variables for multiple curves on the between-individual level
of the second stage. The metadrm() function contains an argument cid2 that
lets us define a curve identifier to group specific individual curves together.
Similar to the argument pmodels in function drm(), the argument pms2 al-
lows us to define different fixed-effects design matrices for each population
parameter on the between-individual level of the second stage.
metaspinach <- metadrm(SLOPE ~ DOSE,
data = spinach,
fct = LL.3(),
ind = CURVE,
cid2 = HERBICIDE,
struct = "UN")
summary(metaspinach)
##
## Two-stage meta-analysis dose-response model
## limit at 0
##
## Call:
## metadrm(formula = SLOPE ~ DOSE, fct = LL.3(), ind = CURVE,
## data = spinach, cid2 = HERBICIDE, struct = "UN")
##
## Variance estimates:
## estim sqrt
## tau^2.1 0.0005 0.0221
## tau^2.2 0.1856 0.4308
## tau^2.3 0.0000 0.0009
##
## rho.b:(I rho.d:(I rho.e:(I
## b:(Intercept) 1 1.0000 -1.0000
## d:(Intercept) 1.0000 1 -1.0000
## e:(Intercept) -1.0000 -1.0000 1
##
##
## Coefficients:
## Estimate Std.Err t value Pr(>|t|)
## b:bentazon 0.5021927 0.0252658 19.8764 9.590e-09 ***
## b:diuron 1.6572431 0.1154678 14.3524 1.654e-07 ***
## d:bentazon 1.3091412 0.2495357 5.2463 0.0005303 ***
## d:diuron 1.9858581 0.3056731 6.4967 0.0001119 ***
## e:bentazon 1.7452319 0.1716524 10.1672 3.116e-06 ***
## e:diuron 0.2086242 0.0091462 22.8100 2.840e-09 ***
## ---
## Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
The correlation estimates at the boundary of the parameter space of −1 and 1

indicate that there might not be enough information available to estimate the
full unstructured random effect covariance matrix Ψ. The struct argument
allows choosing a different correlation structure, e.g., a scaled compound-
symmetry or diagonal structure instead.
The curves for the two different herbicides can be compared by the ratio of
effective dose levels. The EDcomp() function is used to compare the population
estimates of the ED15, ED50, and ED85.
EDcomp(metaspinach,
percVec = c(15, 50, 85),
percMat = rbind(c(1, 1),
c(2, 2),
c(3, 3)),
interval = "delta")
##
##
## bentazon/diuron:15/15 0.75332 0.38388 1.12275
## bentazon/diuron:50/50 8.36543 6.32764 10.40322
## bentazon/diuron:85/85 92.89634 47.22788 138.56480
The herbicide Diuron shows an effective inhibition at much lower dose levels
compared to Bentazon; although, a comparison of relative effective dose levels
is difficult to interpret with large differences between the upper asymptotes.
The same three-parameter log-logistic model can be fitted with the
medrm() function, using a full likelihood approach.
modspinach <- medrm(SLOPE ~ DOSE,

curveid = b + d + e ~ HERBICIDE,
data = spinach,
fct = LL.3(),
random = b + d + e ~ 1|CURVE,
start = c(0.5, 1, 1.5, 1.5, 1.5, 0.3))
round(summary(modspinach$fit)$tTable[, 1:2], 3)
## Value Std.Error
## b.HERBICIDEbentazon 0.503 0.028
## b.HERBICIDEdiuron 1.659 0.087
## d.HERBICIDEbentazon 1.311 0.197
## d.HERBICIDEdiuron 1.986 0.241
## e.HERBICIDEbentazon 1.848 0.205
## e.HERBICIDEdiuron 0.209 0.007
The effective doses can be compared in a similar way as for the two-stage
estimation, resulting in comparable estimates.
EDcomp(modspinach,
percVec = c(15, 50, 85),
percMat = rbind(c(1, 1),
c(2, 2),
c(3, 3)),
interval = "fieller")
##
##
## bentazon/diuron:15/15 0.80056 0.39510 1.23745
## bentazon/diuron:50/50 8.86013 6.90605 10.89971
## bentazon/diuron:85/85 98.05904 63.37850 135.24171
The construction of the plot in Figure 7.1, showing the predicted curves
for each individual assay together with the population prediction is shown in
Section C.3.1.
7.2.3 Herbicides with auxin effects

Four different herbicides were applied in six different concentrations, either as
technical grades material or as commercial formulations to experimental units
consisting of five one-week old seedlings (Streibig, 1987). The data consist of
the 150 observations of dry weights, each observation being the weight of five
plants grown in the same pot. The experiment is replicated three times, each
replication with an additional external control of concentration zero.
2.0
1.5
HERBICIDE
SLOPE
bentazon
1.0 diuron
0.5
0.0
0.01 0.1 1 10 100

DOSE
FIGURE 7.1
Scatterplot with the individual assay and population predictions for the two
herbicides bentazon and diuron.
All the eight herbicide preparations have essentially the same mode of
action in the plant; they all act like the plant auxins, which are plant regulators
that affect cell elongation and other essential metabolic pathways.
head(auxins)
## dryweight dose replicate herbicide formulation
## 1 1.51 0.000 1 control control
## 2 1.43 0.000 1 control control
## 3 0.05 1.000 1 MCPA tech
## 4 0.06 0.500 1 MCPA tech
## 5 0.15 0.250 1 MCPA tech
## 6 0.40 0.125 1 MCPA tech
We assume a three-parameter log-logistic model (Section B.1.1.1) for each

individual herbicide within a replicate. Additionally, we assume that the dry
weight will approach 0 at high concentrations of the herbicide. The variability
between replications is modeled by random replication effects on the upper
asymptote parameter. The effects of each herbicide, formulation, and their
interaction are modeled as fixed-, main-, and interaction effects of the slope,
upper asymptote, and ED50. The external control is joined with the intercept
of the first herbicide, estimating all herbicide and interaction effects to this
reference group.
As a more complex model is needed to analyze this multifactorial de-
sign, we can directly use the function nlme() to fit a nonlinear mixed-effects
model, where the fixed effects are specified by a user-defined design matrix,
copying each predictor variable as a new column into the dataset. The func-
tion model.matrix() automatically creates dummy-coded predictor variables
for the full-factorial combination of herbicide and combination levels. The
user-defined design matrix is constructed by extracting certain predictors and
adding them as new variables to the dataframe.
The order of the treatment levels is changed to define a reference group;
as for the external control, only the information about the upper asymptote
parameter is available.
auxins$formulation <- relevel(auxins$formulation, ref = "tech")

auxins$herbicide <- relevel(auxins$herbicide, ref = "MCPA")
X <- model.matrix(~ herbicide + formulation +
formulation:herbicide, data = auxins)
auxins$h24D <- X[, 3]

auxins$mp <- X[, 4]
auxins$dp <- X[, 5]
auxins$comm <- X[, 7]
auxins$h24Dcomm <- X[, 13]
auxins$mpcomm <- X[, 14]
auxins$dpcomm <- X[, 15]
The fixed effect design matrix Aj is defined by including a list for the argument
fixed, containing a formula for each of the model parameters. The 1 codes
for the combined intercept of the common control together with the reference
group: the herbicide MCPA as a technical grades material. All further terms
denote the effect of each other treatment level to this intercept on the scale
of each model parameter.
The function meLL.3() is the medrc version of the three-parameter log-
logistic model, which is compatible for use in the nlme package environment.
auxmod <- nlme(dryweight ~ meLL.3(dose, b, d, e), data = auxins,

fixed=list(b ~ 1 + h24D + mp + dp + comm +
h24Dcomm + mpcomm + dpcomm,
d ~ 1 + h24D + mp + dp + comm +
h24Dcomm + mpcomm + dpcomm,
e ~ 1 + h24D + mp + dp + comm +
h24Dcomm + mpcomm + dpcomm),
random=d ~ 1|replicate,
start=c(1.2, rep(0,7), 1, rep(0, 7),
0.1, rep(0, 7)))
round(coef(summary(auxmod)), 3)

## b.(Intercept) 1.827 0.287 124 6.370 0.000
## b.h24D 0.443 0.609 124 0.727 0.469
## b.mp -0.244 0.457 124 -0.533 0.595
## b.dp -0.085 0.451 124 -0.188 0.851
## b.comm -0.124 0.520 124 -0.239 0.811
## b.h24Dcomm 0.389 0.957 124 0.406 0.685
## b.mpcomm 0.315 0.729 124 0.431 0.667
## b.dpcomm 0.061 0.734 124 0.084 0.933
## d.(Intercept) 1.278 0.091 124 14.076 0.000
## d.h24D -0.145 0.106 124 -1.364 0.175
## d.mp -0.047 0.148 124 -0.317 0.751
## d.dp -0.069 0.096 124 -0.719 0.473
## d.comm 0.117 0.267 124 0.440 0.661
## d.h24Dcomm -0.081 0.296 124 -0.273 0.785
## d.mpcomm 0.080 0.339 124 0.237 0.813
## d.dpcomm -0.193 0.286 124 -0.674 0.501
## e.(Intercept) 0.065 0.007 124 9.966 0.000
## e.h24D 0.069 0.019 124 3.579 0.000
## e.mp 0.071 0.030 124 2.351 0.020
## e.dp 0.171 0.035 124 4.934 0.000
## e.comm 0.009 0.023 124 0.415 0.679
## e.h24Dcomm -0.014 0.033 124 -0.443 0.659
## e.mpcomm -0.037 0.042 124 -0.880 0.380
## e.dpcomm 0.303 0.082 124 3.683 0.000
The Intercept denotes the estimated slope, asymptote, and ED50 for the
MCPA reference as technical grades material. The h24D, mp, and dp coefficients
show the largest effects on the ED50 scale; all herbicides as technical grades
material reach the ED50 at a higher dose level compared to MCPA. But most
obvious is an interaction effect on the ED50 scale (e.dpcomm), showing that
the commercial formulation of dichlorprop results in an average increase in
ED50 compared to the dp effect as technical grades material.
tech comm
1.5
1.0
herbicide
dryweight
MCPA
24−D
mecorprop
dichlorprop
0.5
0.0
0.02 0.05 0.15 0.4 1 0.02 0.05 0.15 0.4 1

dose
FIGURE 7.2
Scatterplot with the individual replicate and population predictions for four
different herbicides.
The ggplot2 package is used to draw the scatterplot in Figure 7.2, adding
lines for the predictions on the individual replicate and population level. The
R code to construct the plot can be found in Section C.3.2.
7.2.4 Drought stress resistance in Brassica oleracea

The effect of drought stress on Brassica oleracea was investigated by selecting
drought-stess-resistant lines out of a population of different double-haploid
(DH) genotypes (Uptmoor et al., 2009). The study was carried out on 48 DH
lines developed from F1 plants of a cross between the rapid cycling Chinese
kale (Brassica oleracea var. alboglabra (L.H. Bailey) Musil) and broccoli (Bras-
sica oleracea var. italica Plenck). Two stress treatments (not watered and a
watered control) are randomly assigned to four plants per genotype (2 per
treatment) resulting in 192 plants in total. For the genotypes 5, 17, 31, and
48, an additional 12 plants (6 per treatment) are included in the completely
randomized design, which results in a total of 240 plants. For each plant, the
length of the youngest leaf at the beginning of the experiment is measured
daily for a period of 16 days. For the additional 12 plants of the 4 genotypes,
the leaf water potential was measured as a secondary endpoint (omitted here);
due to these destructive measurements, some dropouts occurred.
The growth curves are assumed to follow a three-parameter logistic model
(Section B.1.1.1), setting the lower asymptote to zero. Population parameters
are defined as an intercept at the control treatment and as the difference in
treatment levels when stressing the plants. For each genotype, an individual
intercept and stress effect is assumed as random effects with an unstructured
covariance matrix. The correlation between repeatedly observed leaf lengths
within a plant is estimated assuming an autoregressive AR1 structure for the
residuals.
brmod <- nlme(LeafLength ~ meL.3(Day, b, d, e), data = broccoli,
fixed=b + d + e ~ 1 + Stress,
random=b + d + e ~ 1 + Stress | Genotype,
correlation = corAR1(form = ~ Day | Genotype/ID),
start = c(-0.4, 0, 15, 0, 5, 0))
summary(brmod)
## Nonlinear mixed-effects model fit by maximum likelihood

## Model: LeafLength ~ meL.3(Day, b, d, e)
## Data: broccoli
## AIC BIC logLik
## 4444.401 4624.581 -2193.2
##
## Random effects:
## Formula: list(b ~ 1 + Stress, d ~ 1 + Stress, e ~ 1 + Stress)
## Level: Genotype
## Structure: General positive-definite, Log-Cholesky
## parametrization
## StdDev Corr
## b.(Intercept) 0.01370399 b.(In) b.Strs d.(In) d.Strs e.(In)
## b.Stressdrought 0.03609302 0.322
## d.(Intercept) 1.86995266 -0.469 0.123
## d.Stressdrought 1.04427398 0.595 -0.252 -0.839
## e.(Intercept) 0.47151685 0.307 0.306 0.562 -0.517
## e.Stressdrought 0.46963336 0.463 0.196 -0.144 0.574 -0.240
## Residual 1.24750262
##
## Correlation Structure: ARMA(1,0)
## Formula: ~Day | Genotype/ID

## Parameter estimate(s):
## Phi1
## 0.9484194
## Fixed effects: b + d + e ~ 1 + Stress
## b.(Intercept) -0.366859 0.00626349 3636 -58.57106 0
## b.Stressdrought -0.067694 0.01163874 3636 -5.81630 0
## d.(Intercept) 14.044417 0.29914816 3636 46.94803 0
## d.Stressdrought -3.267275 0.22960524 3636 -14.22997 0
## e.(Intercept) 5.268797 0.09833708 3636 53.57894 0
## e.Stressdrought -1.442174 0.12301961 3636 -11.72312 0
## Correlation:
## b.(In) b.Strs d.(In) d.Strs e.(In)
## b.Stressdrought -0.433
## d.(Intercept) 0.076 -0.060
## d.Stressdrought -0.149 0.254 -0.734
## e.(Intercept) 0.060 0.103 0.369 -0.261
## e.Stressdrought 0.098 0.020 -0.085 0.222 -0.495
##
## Standardized Within-Group Residuals:
## Min Q1 Med Q3 Max
## -5.59627518 -0.58080992 -0.03044674 0.58513286 4.72865998
##
## Number of Observations: 3689
## Number of Groups: 48
The summary output shows the estimated population parameters as fixed ef-
fects for the control group (intercept), and the difference between parameters
for drought stress plants compared to the control. The estimated negative
population effects indicate a slower growth with a smaller slope and the in-
flection point being located at an earlier time. Additionally, the estimated
effect for the asymptote shows that the drought stress treatment results in
smaller plants. On the scale of each of these parameters, the between-genotype
variability is estimated; the genotype standard deviation is presented in the
Random effects section together with the pairwise correlation coefficients.
The dependency between residuals from a single plant is parametrized as the
coefficient of the autoregressive AR1 structure. The estimate of the coefficient
is near one, indicating highly correlated residuals within a plant.
Additionally, the predicted random effects can be obtained with the func-
tion ranef() or alternatively the linear function of fixed and random effects
with the function coef() (not shown here). Instead, we can visualize the
genotype-stress interaction by plotting the predicted leaf lengths on the level
of each genotype-stress combination in Figure 7.3. The corresponding R code
can be found in Section C.3.3.
1 2 3 4 5 6 7
20
15
10
5
8 9 10 11 12 13 14
20
15
10
5
15 16 17 18 19 20 21
20
15
10
5
LeafLength
22 23 24 25 26 27 28 Stress
20
15
10 control
5
drought
29 30 31 32 33 34 35
20
15
10
5
36 37 38 39 40 41 42
20
15
10
5
4 81216
43 44 45 46 47 48
20
15
10
5
4 81216 4 81216 4 81216 4 81216 4 81216 4 81216
Day
FIGURE 7.3
Plant-specific growth curve predictions, comparing drought-stressed broccoli
plants vs. a watered control for each genotype.
The drought stress mainly has an effect on the upper asymptote parameter,
that is, the stressed plants are smaller than the watered plants. But we can also
see a large negative correlation between the genotype-specific stress effects and
the genotype-specific asymptote at the control; hence, genotypes with larger
leaves also show the larger response to the drought stress treatment.
Appendix A
Estimation
Once a suitable dose-response model function has been found, the next step
is to choose a suitable estimation procedure, which should ideally exploit the
type of response as much as possible.
Least squares estimation, which is a special case of maximum likelihood
estimation, should be used for continuous response variables that are approx-
imately normally distributed with the same standard deviation for all doses
(variance homogeneity) (Meister and den Brink, 2000). Section A.1 below
provides more details. More general maximum likelihood estimation involv-
ing distributional assumptions different from the normal distribution, may be
more appropriate in case the response is binary, a sum of binary variables
(Piegorsch and Bailer, 2005, pp. 172-179), or a count with a substantial por-
tion of zeros (Kerr and Meador, 1996; Ritz and Van der Vliet, 2009). More
details are provided in Section A.2 below. In practice, least squares estimation
is often applied also for responses that are not normally distributed. This ap-
proach makes sub-optimal use of the information available in the data, often
resulting in a loss in efficiency seen as unnecessary large standard errors (Szöcs
and Schäfer, 2015). Therefore, it is important to ensure alignment between the
type of response and the estimation procedure used.
To deal with model misspecification the results from maximum likelihood
estimation or even the entire estimation procedure may be modified (see Sec-
tion A.3 on transformations, Section A.4 on robust estimation, and Section A.5
on sandwich variance estimators).
Constrained estimation may be needed in case certain restrictions have to
be enforced for certain parameters (see Section A.6). However, surprisingly,
in most cases unconstrained estimation will suffice, i.e., no bounds need to be
imposed on ranges of any parameters even though ranges for some parameters
may be restricted (e.g., ED50/LD50 has to be non-negative), as usually reason-
able parameter estimates will be obtained if data carry sufficient information
about the parameters. Unreasonable parameter estimates often imply lack of
information. It can also happen that the estimation procedure is not successful
in finding the optimal parameter estimates. For dose-response analysis such
problems often happen due to poor choice of starting values for the parameter
estimates: the estimation procedure needs to be provided with good starting
values when searching for the optimal parameter estimates. The package drc
relies heavily on self-starter routines that compute data-driven starting values.
Section A.8 explains the idea in more detail.
161
Estimation of the model parameters or any derived parameters utilizes the

entire dose-response data. For some specific parameters, observations may still
contribute in an unequal way, e.g., for a decreasing dose-response relationship
the estimated upper limit at dose 0 will be mostly influenced by the low-
dose range response values, although all response values will contribute to
the corresponding standard error (at least in case variance homogeneity is
assumed). The uncertainty on parameter estimates is described by means of
standard errors or confidence intervals (see Section A.9).
For dose-response analysis, it is of particular interest to estimate effective
doses through inverse regression: assuming a given response value and then
deriving the corresponding mean dose value. This is the opposite operation
of prediction where a mean response is predicted for a given dose. There are
various ways to estimate effective doses and the corresponding standard errors.
Section A.10 provides details on the different approaches implemented in drc.
A.1 Nonlinear least squares

Assume that x1 , . . . , xn are the dose values and y1 , . . . , yn the corresponding
observed responses (assuming a sample size of n). The nonlinear least squares
estimates are obtained by minimizing the following sum of squares:
n
X
wi2 {yi − f (xi , β)}2 (A.1)
i=1
where wi ’s are user-specified weights (often left unspecified, i.e., equal to 1).
β is the vector of all model parameters, i.e., the parameters b, c, d, . . . of
Appendix B.
In drm(), weights are specified through the argument weights and they
should be on the same scale as the response, e.g., expressed as standard de-
viations and not empirical variances. Weights may be used for addressing
variance heterogeneity in the response. However, the transform-both-sides ap-
proach should be preferred instead of using often very imprecisely determined
weights.
Equation (A.1) has to be solved numerically in an iterative manner. One
approach is to use iteratively weighted least squares as is done in nls(), which
is part of the standard installation of R (Ritz and Streibig, 2008). Another
approach is to use a general-purpose minimizer directly, as in drm() where
optim() is used in combination with some pre-scaling of parameters. In our
experience drm() is more robust than is nls(); lack of robustness of nls()
has also been pointed out elsewhere (Nash, 2014).
Estimation 163
The estimated variance-covariance of the parameter estimates (cov(β̂))

is obtained as the scaled inverse of the observed information matrix, which
consists of second-order partial derivatives of f with regard to the model
parameters β1 , . . . , βp :
!−1 −1
2
∂2f

2 ∂ f 2
cov(β̂) = σ̂ = σ̂ (A.2)
∂βp1 ∂βp2 p1 ,p2 ∈{1,...,p} ∂βp1 ∂βp2
The scaling factor σ̂ is the residual standard error, which is estimated in the
same way as in linear regression. The observed information matrix (“hessian”)
in equation (A.2) is approximated numerically in optim() upon convergence.
However, it is not even a requirement that the inverse is available: if it is not
then NAs are returned for standard errors.
A.2 Maximum likelihood estimation

In order to obtain parameter estimates, the following expression has to be
maximized in terms of β values in order to obtain parameter estimates:
n
X
log likelihood = l{yi , f (xi , β)} (A.3)
i=1
where l denotes the logarithm-transformed likelihood function for a single

observation (i.e., we look for the β values that render the value of this function
as large as possible). The likelihood is entirely defined by the data and the
model assumptions, i.e., distributional assumptions and dose-response model
function. The log likelihood value for a given dose-response model fit may be
obtained using the function logLik().
Estimation based on the maximum likelihood principle provides approx-
imate estimated standard errors and, consequently, approximate confidence
intervals. In particular, estimated standard errors are obtained from the esti-
mated variance-covariance matrix of the parameter estimates, which is calcu-
lated (approximated numerically) as the inverse of the observed information
matrix based on second derivatives of the log likelihood function (van der
Vaart, 1998):
−1
∂2l

cov(β̂) = (A.4)
∂βp1 ∂βp2
The estimated variance-covariance matrix is available through the function
vcov(). The estimated standard errors are the square root values of the di-
agonal elements in this matrix.
Likelihood ratio tests can be used for comparing nested models. Moreover,
information criteria such as AIC and BIC can also be calculated and used
for comparing non-nested models such as comparing different choices of the

model function F .
As detailed below, maximum likelihood estimation is used for fitting dose-
response models to binomial, count, and time-to-event data, including models
for binomial data where natural immunity and/or natural mortality is present
(Finney, 1971).
A.2.1 Binomial dose-response data

A binary response can take on values 0 or 1. A binomial response is the sum
of a fixed number of binary responses (the number of 1’s). Specifically, let Yi
denote a binomial response obtained in a dose-response experiment while a
dose xi was applied. Then the statistical model may be specified as follows:
E(Yi ) = nf (xi , β)
p
SD(Yi ) = nf (xi , β)(1 − f (xi , β)) (A.5)
Thus, the standard deviation of a binomially distributed response depends
solely on the number of basic binary experiments and on the assumed dose-
response model function.
A.2.2 Count dose-response data

A.2.2.1 The Poisson distribution
Let Yi denote the count observed corresponding to the dose xi . Then the
statistical model may be specified as follows:
E(Yi ) = f (xi , β)
p
SD(Yi ) = f (xi , β) (A.6)
Thus, the standard deviation of a Poisson distributed count is equal to the
square root of the mean of the count. For√instance, counts√with mean 10
will most of the time lie between 10 − 2 · 10 and 10 + 2 · 10, that is in
the interval [4, 16]. Likewise, counts with mean 100 will most of the time lie
between [80, 120]. Therefore the variation of a Poisson distributed count is
somewhat constrained and it will often not suffice to describe the extent of
variation observed in reality in count data. Such excess variation compared
to the Poisson distribution is usually referred to as over-dispersion (Morgan,
1992, Chapter 6).
Weights may also be incorporated in the Poisson model:
E(Yi ) = f (xi , β) · wi (A.7)
where wi denotes the weight associated with the count Yi (Delignette-Muller
et al., 2014b). As before the standard deviation is equal to the square root of
the mean.
Estimation 165
A.2.2.2 The negative-binomial distribution

The Poisson model assumes that the standard deviation is equal to the square
root of the mean. However, when dealing with count data, it is not unusual
to observe larger variation than predicted by the Poisson model. One way to
deal with such over-dispersion is to turn to the negative binomial distribu-
tion, which has an additional parameter that allows the standard deviation
to differ from the square root of the mean. Specifically the negative-binomial
distribution implies the following mean and standard deviation:
E(Yi ) = f (xi , β)
p
SD(Yi ) = f (xi , β) + ω(xi , β)2 (A.8)
where the parameter ω ≥ 0 reflects the degree of over-dispersion in the data:

the value 0 corresponds to the Poisson model (no overdispersion).
A.2.3 Time-to-event-response data

The statistical model corresponds to assuming that probabilities of an event
happening in the different monitoring intervals are described by a multinomial
distribution (McCullagh and Nelder, 1989, pp. 164–184, Jager et al., 2011),
with probabilities determined by the assumed dose-response model. Specifi-
cally, for each monitoring interval ]tj−1 , tj ] the estimation procedure will seek
to match optimally the corresponding observed proportions of events nj /n
against the probabilities of the event happening. More specifically, the like-
lihood function, which is maximized as a function of the model parameters
in order to obtain the parameter estimates, is defined as the product of the
probabilities of events happening in the different monitoring intervals, based
on all data. In practice, maximization of the logarithm-transformed likelihood,
which results in the same estimates, is more computationally convenient:
q
X
l= nj log(F (tj ) − F (tj−1 )) (A.9)
j=1
Monitoring intervals where no event was observed may be left out of the
analysis as they will not contribute to the estimation (nj = 0 implies a zero
term in the log likelihood); in some cases they have to be left out to achieve
convergence of the estimation procedure.
The implicit assumption of the above log likelihood is that right-censoring
only happens at the end of the experiment, i.e., it is possible to follow all
seeds or organisms until the end of the experiment unless the event of interest
happens. However, it could also happen that some seeds or organisms are
right-censored during the experiment, e.g., seeds are dormant. Thus, the log
likelihood would need to be modified slightly (Ritz et al., 2013b).
Model checking can be done by visually assessing the agreement between
the observed cumulative germination curve and the fitted curve based on, or
by looking at, the residual plot based on cumulative residuals (McCullagh and
Nelder, 1989, p. 179).
A.3 The transform-both-sides approach

The transform-both-sides approach implies that both the response and the
nonlinear model function are shifted by a constant C and then transformed
by a suitable function gλ (Carroll and Ruppert, 1984). The resulting model
for the mean of the transformed response looks like this:
E{gλ (y + C)} = gλ (f (x, (β1 , . . . , βp )) + C) (A.10)
with the transformed response on the left-hand side and the transformed
model function on the right-hand side. Usually the function gλ is taken to
be the Box-Cox transformation gλ (y) = (y λ − 1)/λ for some suitable choice of
λ ∈ R. The value λ = 1 implies no transformation whilst λ = 0 corresponds
to the logarithm transformation. All other values of λ correspond to power
transformations. It is noteworthy that the Box-Cox transformation may al-
leviate variance heterogeneity and some skewness in the distribution of the
response and thus recover a normal distribution, but it may not remedy other
problems with the distributional assumptions such as counts observed with
ties (Ritz and Van der Vliet, 2009).
The package drc provides a boxcox method, which has been implemented
in much the same way as the corresponding method for linear models available
in the package MASS (Venables and Ripley, 2002). There is, however, a choice
between a profiling approach as used for linear models in MASS or a more
robust analysis of variance (ANOVA) approach where the optimal λ is esti-
mated from a more general ANOVA model, i.e., a linear model, and not from
the specified dose-response model; the latter requires replicate observations
for at least some doses.
Manual specification of the transformation is also possible through the
arguments bcVal and bcAdd, which correspond to λ and C in equation (A.10),
respectively. Using a specific transformation (i.e., a specific value of λ and
possibly C) based on previous experience should be preferred over using a
data-driven choice.
A.4 Robust estimation

One way to handle observations that seem to be extreme or deviating in
one way or another as compared to the bulk of the observations is to apply
Estimation 167
a robust estimation procedure, which will weigh down the influence of such
observations.
In drc robust dose-response analysis with a continuous response is available
through the argument robust using the same models as considered in the case
of robust linear regression (Venables and Ripley, 2002), except for the model
based on Hampel’s ψ, which is currently not implemented. However, it may
be difficult to fit such robust nonlinear models unless accurate starting values
for the model parameters are provided.
The estimated “information”-type variance-covariance has the following
form (Huber, 1981; Stromberg, 1993):
−1
∂2ρ

2
cov(β̂) = σ̃ (A.11)
∂βp1 ∂βp2
where the function ρ controls the influence of observations on the estimation
procedure (e.g., ρ(y) = y 2 corresponds to ordinary nonlinear least squares
estimation). Once parameter estimates have been obtained, all methods and
extractors available in drc may be used as if the model fit had been obtained
using maximum likelihood estimation.
A.5 Sandwich variance estimators

Consider the situation where a dose-response model has been chosen with
mean function f , which is correctly specified, but the assumptions regarding
normality and/or the variance homogeneity are not satisfied. In this case, the
estimated standard errors from the model assuming normality and variance
homogeneity will be inconsistent: They will not approach the true standard
errors as the sample size increases. However, even though the model is mis-
specified with regard to some of the distributional assumptions, it is still
possible to obtain consistent estimates of the standard errors by adjusting the
estimated variance-covariance matrix (Carroll and Ruppert, 1988, p. 128).
In case of replicates, one way to calculate robust standard errors is to
imagine that the standard error of a predicted response corresponding to a
dose included in the experiment is being scaled up or down according to the
magnitude of residuals for that particular dose relative to the residual standard
error. So there can be different scaling at work for different doses. We give the
mathematical definition below.
The modified variance-covariance matrix is defined below.
c β̂) = s2 B̂ M̂ B̂
var( (A.12)
The matrix M̂ is a function of the first derivatives of the log likelihood
function and s2 B̂ is the usual variance-covariance matrix. This equation is
valid as long as the mean structure is correct and independence can be as-
sumed. Therefore, the resulting estimated variance-covariance matrix is said
to be robust against misspecification of the distributional assumptions. If the

model is correctly specified, we have the equality M̂ = B̂ −1 , and the mod-
ified variance-covariance matrix reduces to the usual model-based variance-
covariance matrix.
The modified variance-covariance matrix is often called a sandwich vari-
ance estimator due to the particular product form on the right-hand side
of Equation (A.12). The resulting standard errors are referred to as robust
whereas the usual estimated nonrobust variance-covariance matrix results in
naı̈ve standard errors. More detailed explanations can be found in White
(1981), Carroll and Ruppert (1988, pp. 209–213), White (1996, Chapter 5),
and Zeileis (2006).
The use of sandwich variance estimators is attractive because we avoid
having to find a transformation or specify a variance model but can still adjust
for deviations from the model assumptions. The sandwich variance estimators
are obtained in drc by utilizing the packages lmtest and sandwich (Zeileis and
Hothorn, 2002; Zeileis, 2006).
A.6 Constrained estimation

By default, unconstrained estimation is carried out by drm(), although for
most models there are actually constraints on at least some of the model
parameters, e.g., ED50 needs to be positive. Likewise, there may also be con-
straints needed to ensure uniqueness of the model fits, e.g., the slope parame-
ter b should be negative for an increasing dose-response relationship to ensure
that c is the lower asymptote and d the upper asymptote. In practice, such
constraints may often be indirectly enforced through the appropriate choice
of starting values.
However, in principle, any of the aforementioned estimation procedures
may be combined with the use of constraints where the range of one or more
parameters is restricted, e.g., to certain intervals by setting lower and upper
bounds (so-called box constraints) that are different from −∞ and ∞, respec-
tively. This is possible with drm() using the arguments lowerl and upperl,
which, if specified, will invoke constrained optimization through the method
“L-BFGS-B” in optim().
A.7 Two-stage estimation for hierarchical models

Estimating the unknown parameters in the hierarchical nonlinear model by
maximum likelihood is a challenge, as the likelihood function includes an im-
proper integral, that is, does not have a closed-form solution (Demidenko,
Estimation 169
2013). Several estimators for β are available, where the most popular among
R users utilizes the Lindstrom-Bates algorithm.
A.7.1 Technical replicates

If there is large uncertainty about each observation for an individual, e.g.,
using a device that exhibits large variability for each measurement, this addi-
tional source of variance can be represented by observing multiple technical
replicates from the same experimental unit. A simple two-stage analysis strat-
egy can be applied, calculating an average response for each individual and
assuming a dose-response model for the estimated averages in a second stage.
It is obvious that this approach only works for repeated measurements at a
single dose level within an individual. As the uncertainty about estimating
each average is ignored in the second stage, there is some loss of informa-
tion as each average obtains the same weight. Hence, the same number of
observations is required for each individual.
A.7.2 Two-stage approaches

If an experimental design is planned with repeated measurements for differ-
ent dose levels within an individual, the two-stage approach for the technical
replicates can be extended to a more general framework (Demidenko, 2013).

1. In a first step, a separate curve f j xj , β j is fitted for each of the
M individuals, using nonlinear least-squares. For each of the M
models we obtain a set of coefficient estimates β̂ j together with the
corresponding variance estimates.
2. Estimates for the population parameters can be obtained in a sec-
ond step, treating the estimated coefficients for each individual as
a new response in a linear model.
β j ≈ Aj β̃ + B j bj + ?j
The first stage requires that each individual curve can be fitted; that is,
the number of observations need to be larger than the number of unknown
parameters and the full dose range should be covered. Further, it requires that
there is no common parameter shared between individual curves; otherwise,
it would not be possible to fit the curves separately from each other in the
first stage.
For the estimation of the population parameters in the second step, a
method-of-moments estimator of the between-individual variance Ψ can be
obtained, followed by a generalized least-squares estimation of β̃ (Demidenko,
2013). Alternatively, β̃ and Ψ can be estimated by maximum likelihood (Da-
vidian and Giltinan, 1995, Chapter 5.3), incorporating the uncertainty about
estimating the individual β̂ j by a residual term ?j . The individual residual
variances are fixed during the maximization of the likelihood in the second
step.
The concept of the two-stage approach can also be presented as a meta-

analysis framework (Jiang and Kopp-Schneider, 2014; Viechtbauer, 2010). As-
suming that the individual curves originate from separate experiments, one
can combine the estimated coefficients by a weighted average. The weights
depend on the residual variance of the individual curves and the between-
experiment variance. Again, several methods of moments and maximum like-
lihood estimators are available to obtain the heterogeneity estimators (Kalaian
and Raudenbush, 1996; Berkey et al., 1998; van Houwelingen et al., 2002).
A.7.3 Lindstrom-Bates algorithm

The Lindstrom-Bates algorithm is a two-stage procedure, approximating the
nonlinear model by a linear mixed-effects model (Lindstrom and Bates, 1990;
Pinheiro and Bates, 2000).
1. In the first step, a penalized least-squares algorithm is applied, min-
imizing
XM
ky j − f j xj , β j k2 + k∆bj k2

j=1
to find estimates of the fixed-effects parameters while holding the

precision factor ∆ fixed, which factors the precision matrix of the
random effects and expresses it relative to the precision of the resid-
uals.
2. In the second step, a linear mixed-effects model is applied to a vector
of weights as a new response and design matrices X̂ j and Ẑ j for
the fixed and random effects, which are obtained at the parameter
estimates from the first step by
ˆ + Ẑ b̂

ŵj = y j − f j xj , β̂ j + X̂ j β̃ j j
∂f j ∂f j
X̂ j = 0 ˆ , Ẑ j =
∂b0j β̃,
ˆ
∂ β̃ β̃,b̂j b̂j
The precision factor ∆ that is estimated in the second step is again plugged
in the penalized least-squares step. The algorithm then iterates between the
two steps until convergence. Pinheiro and Bates (2000) and Demidenko (2013)
provide additional information about the algorithm and numerical details of
its implementation in R.
Following Pinheiro and Bates (2000), we can assume the following distri-
bution for the fixed-effects estimator:
  −1 
M
ˆ ∼ N β̃, σ 2 X X̂ Σ−1 X̂   , where Σ = I − Ẑ ∆−1 ∆−T Ẑ T
β̃  j j j  j j j
j=1
Estimation 171
A.8 Starting values and self-starter functions

Due to the nonlinearity of most dose-response models, a critically important
aspect of estimation in dose-response analysis is the choice of the so-called
starting values for the model parameters. The numerical optimization proce-
dure used for estimating the model parameters needs to be initialized at some
values of the model parameters from which it can iteratively approach the
optimal values as judged by some criterion. The optimal values of the model
parameters are the parameter estimates, which we interpret as providing the
best-fitting dose-response curve(s) to the data.
Although the starting values may be derived from the same dose-response
data for which a dose-response model has to be fitted, the starting values are
not to be understood as parameter estimates and they cannot be used for
drawing any kind of statistical inference. They are simply a tool for getting
the estimation to work.
The estimation procedures used for dose-response analysis in this book are
iterative and they need to be initiated by starting values for the model param-
eters. The choice of starting values may crucially affect whether convergence
is eventually achieved. Thus, availability of good starting values facilitates
parameter estimation in nonlinear models.
Starting values may be obtained either by using parameter estimates pre-
viously reported for similar experiments or, in a data-driven way, by using
the dose-response data themselves to elicit relevant information. For instance,
within weed science, routine calculation of starting values for log-logistic mod-
els from simple linear regression models of transformed responses dates back at
least to Streibig et al. (1993), utilizing extended linearization techniques previ-
ously used for probit analysis and modified for nonlinear regression (Streibig,
1981). The flexibility of R has made it possible to implement so-called self-
starter functions that return data-driven starting values for the model param-
eters. In R, self-starter functions are available for the function nls(). In drc
the above-mentioned data-driven linearization technique has been extended
to other types of dose-response models and it is now available for all built-
in drc models. Use of the self-starter function is the default setting, which
may, however, be overruled by supplying starting values manually through
the argument start. The function getInitial() may be used to obtain the
starting values that were actually used for obtaining a particular model fit.
If a dose-response model is fitted to dose-response data with multiple
curves, then starting values are needed for all model parameters in all groups.
The parameters should be provided for each model parameter, i.e., the val-
ues for the parameter b first, next the one for the parameter c, and so on in
alphabetic order.
More sophisticated self-starter functions have been proposed (Normolle,
1993), but so far they have not been implemented in drc except for log-logistic
models.
A.9 Confidence intervals

For a dose-response model that assumes normally distributed response values,
the confidence interval of a parameter is defined as the corresponding param-
eter estimate ± K · estimated standard error of the parameter estimate. To
obtain a 95% confidence interval, the constant K is equal to the 97.5% per-
centile in a t distribution with degrees of freedom equal to the number of
response values minus the number of parameters in the model function used
in the model.
For all other types of dose-response data, Wald-type confidence intervals
are used, i.e., to obtain a 95% confidence interval, the constant K is equal to
the 97.5% percentile in the standard normal distribution (K = 1.96). These
intervals are often referred to as Wald intervals.
By definition, these confidence intervals are symmetric around the esti-
mate, which occasionally may result in lower bounds that are unrealistic, e.g.,
a negative lower limit for the confidence interval for ED50 or for the con-
fidence band for the entire dose-response curve. There are several ways to
ensure that they fall within a realistic range. One simple approach is to trun-
cate lower limits and bounds at 0 (Feng and McCulloch, 1992); this approach
is implemented in the predict() method.
A pointwise confidence band for the dose-response curve is a collection of
confidence intervals corresponding to doses in a given interval. The band is
pointwise as 95% coverage is only established for each confidence interval, but
not for the entire confidence band, which will have somewhat less than 95%
coverage.
Inverse regression of the confidence band of the dose-response function
also leads to asymmetric confidence intervals for ED values (Nottingham and
Birch, 2000). For ED values too close to 0 or 1 the resulting confidence intervals
may degenerate, having either 0 as the lower limit or ∞ as the upper limit.
A.10 Prediction and inverse regression

There are a number of functions available in drc for obtaining predicted val-
ues from a dose-response model fit. Notably, there are the general-purpose
methods fitted() and predict(). The latter may also provide confidence
or prediction intervals (controlled through the argument type). The function
backfit() may be used for obtaining the estimated doses that correspond to
the observed average response at each dose.
Estimating doses corresponding to specific response levels is often of partic-
ular interest in dose-response analysis. Such estimates are obtained by solving
an inverse regression problem, which may be approached either by means of
Estimation 173
after-fitting or using a re-parameterization. After-fitting means solving the

inverse regression problem after having fitted the model. For some models
the solution may be derived explicitly (e.g., log-logistic models (Streibig and
Jensen, 2000; Ritz and Streibig, 2005)) whereas for other models only numer-
ical solutions are available, e.g., hormesis models (Cedergreen et al., 2005).
Re-parameterization means fitting the model again using a reparameterized
model function where the parameter of interest is a model parameter.
After-fitting relies on the so-called delta method, which is a statistical
method for obtaining approximate standard errors for derived parameter es-
timates, i.e., parameter estimates that are functions or transformations of the
estimated model parameters. The intuitive explanation is as follows: For a
specific function g(β) of the model parameter β, the delta method utilizes
mathematical calculus (first-order differentiation of g in β) to derive a scaling
factor that will convert the estimated standard error of β̂ into an estimated
approximate standard error of the derived parameter estimate g(β̂). It works
if it is reasonable to assume local linearity of the function g in the neigh-
bourhood of β̂; it can be a nonlinear function, but it should not fluctuate too
rapidly. For a more detailed and technical explanation, we refer to Weisberg
(2005).
A practical implication of the after-fitting approach is that it suffices to
fit a dose-response model once, in a parameterization that has proven to be
the most stable for estimation. In our opinion, after-fitting provides a major
improvement over the often quite unstable re-parameterization approach.
A.10.1 Effective dose

The most prominent example of a derived parameter obtained using inverse
regression is the dose that results in a halfway reduction between the lower
and upper limits of the dose-response curve f as doses approach infinity and
0, respectively. This parameter is often denoted ED50 or EC50 for continuous
responses, and LD50 or LC50 for binomial responses, and t50 for event-time
responses.
For a monotonously decreasing mean function f , the effective dose for
0 < α < 1 (ED100α) is defined as the solution to the following equation:
f (ED100α, β) = (1 − α) lim f (x, β) + α lim f (x, β) = (1 − α)c + αd (A.13)

x→0 x→∞
Note that these limits are well defined and finite for dose-response models (for
most models in drc they correspond to the parameters d and c, respectively).
For a monotonously increasing mean function the same equation is obtained
by interchanging the limits.
The resulting effective dose is a relative quantity, defined in terms of a
percentage reduction. For instance, ED50 (α = 0.50) is the dose resulting
in a 50% reduction in the average response relative to the lower and upper
limits of f . Such relative effective doses are mostly suitable for continuous
responses. For increasing α, the corresponding ED100α will be increasing as

a consequence of f being a decreasing function.
For binomial responses, absolute effective doses referring to the entire prob-
ability scale [0, 1], which do not necessarily coincide with the lower and upper
limits of the estimated dose-response curve, are usually more relevant. In gen-
eral, an absolute effective dose EDy0 corresponding to an average response y0
may be defined as the solution to the following equation:
f (EDy0 , β) = y0 (A.14)
An absolute effective dose for y0 ∈]c, d[ may always be calculated as some rel-
ative effective dose for a suitably derived α value. In some cases, this approach
will involve parameter estimates of the lower and upper limits, but at present
the variation in these estimates will not be propagated to the estimated effec-
tive dose.
Estimated effective doses are obtained by inserting parameter estimates
and solving equation (A.13) with regards to the derived parameter ED100α.
In drc, the function ED() will calculate estimated effective doses, and the
argument type controls the type of effective dose being calculated: relative
(default) or absolute. Note also that for hormesis models, effective doses
may also be meaningfully defined for some α < 0 (by setting the argument
bound=TRUE in ED()). Model-averaged estimated ED values may be obtained
through the function maED() (Kang et al., 2000).
A.10.2 Relative potency

The ratio of two effective doses ED100α1 and ED100α2 (0 < α1 < 1 and
0 < α2 < 1) of two fitted dose-response curves denoted A and B, say, is
defined as follows:
EDA 100α1
ρ(α1 , α2 ) = , 0 < α1 , α2 < 1
EDB 100α2
For the important special case where α1 = α2 , the ratio is referred to as
the relative potency (Finney, 1971), and it is interpreted as a measure for
quantifying the strength of one substance over another, i.e., a pairwise com-
parison based on a ratio. The term selectivity index is also sometimes used
(Christensen et al., 2003). In general, the ratio may be interpreted as the
order of magnitude of the window between harmful and safe for appropriate
choices of α1 and α2 , e.g., 0.1 and 0.9. In some special cases (“parallelism”)
the relative potency is constant between two dose-response curves for all val-
ues satisfying α1 = α2 . More often it will change with the value of α1 = α2
and this can be investigated using the function relpot() (Ritz et al., 2006).
In drc the extractor function EDcomp() may be used to obtain estimated
relative potencies or differences between effective doses. By default the corre-
sponding standard errors are derived using the delta method and confidence
intervals are based on these standard errors (indicated through the argument
Estimation 175
interval = "delta") in combination with the t- or z-value (a percentile in

an appropriate t- or normal distribution). Alternatively, confidence intervals
based on the approach of Fieller may be used (interval = "fieller") (Mor-
gan, 1992; Faraggi et al., 2003). For the special case of the log-logistic model
where log(ED50 ) is a model parameter (ẽ), confidence intervals on the orig-
inal dose scale may be retrieved through back transformation (interval =
"fls").
Appendix B
Dose-response model functions
In this chapter, we provide detailed descriptions of dose-response models be-

longing to four main classes of models: log-logistic, log-normal, Weibull type
1, and Weibull type 2 models. These four classes of models may be seen
as direct extensions of the four types of generalized linear models proposed
by McCullagh and Nelder (1989, pp. 108–109) for analysis of binomial data,
corresponding to the logit, probit, log-log, and complementary log-log link
functions, respectively. The extensions include more model parameters and
are useful for various types of dose-response data, not only binomial dose-
response data. Some other types of dose-response models are also introduced.
Table B.1 and Table B.2 provide an overview of the models available in drc.
However, this chapter does not provide an exhaustive account of all available
dose-response models.
Many of these dose-response model functions, which we describe below,
may be constructed via a cumulative distribution function (cdf) or some other
suitable function F , say, in the following way:
f (x) = c + (d − c)F (x, b, . . .) (B.1)
(Seber and Wild, 1989, p. 337). The parameter c denotes the lower asymptote
or limit of the dose-response model, and the parameter d denotes the upper
asymptote or horizontal limit of the dose-response curve. There will always be
a parameter b, which, in one way or another, will reflect the rate of change of
the dose-response curve between the upper and lower limits, analogous with
the slope coefficient of a simple linear regression model. However, in general,
dose-response models have no equivalent to an intercept and hence there is
no model parameter named a. The number of additional parameters (denoted
. . .) will depend on the choice of F (usually 1–3) and these parameters will be
denoted e, f etc.
In addition, many of the model functions are scale invariant in the sense
that the model itself, usually through the parameter e, which acts as a scaling
factor, accommodates the magnitude of doses. Likewise, many of the model
functions involve the logarithm. We use the natural logarithm and its inverse,
the exponential function (power of e), but most models may also be formulated
using any other logarithm (resulting in a slightly different reparameterization
of the models). For instance, models using the base-10 logarithm are occasion-
ally used.
177
B.1 Log-logistic models

The log-logistic models are by far the most commonly used dose-response mod-
els (e.g., Environment Canada, 2005; Van der Vliet and Ritz, 2013). The basic
model may be derived from the cdf of the standard log-logistic distribution.
Sometimes log-logistic models are confusingly called logistic models (e.g.,
Ritz and Streibig, 2005). Strictly speaking, logistic models, however, constitute
a different class of models and are in general not as suitable for describing dose-
response data as they are defined for the entire real axis and not only for the
non-negative dose scale (Ritz, 2010). Logistic models may, however, be useful
in other contexts where the x axis is not confined to the non-negative part.
B.1.1 Four-parameter log-logistic models

The four-parameter log-logistic model corresponds to the following model
function:
d−c
f (x, (b, c, d, e)) = c +
1 + exp(b(log(x) − log(e)))
d−c
=c+ = c + (d − c)(1 + (x/e)b )−1 (B.2)
1 + (x/e)b
The four parameters are b, c, d, and e. The parameter e, which by definition

has to be positive, corresponds to the dose producing a response half-way
between the upper limit d and lower limit c, i.e., the parameter e is the ED50.
The slope parameter b reflects the steepness of the dose-response curve at the
dose ED50. Specifically, the relative slope of the log-logistic model function
has the opposite sign as compared to the sign of the parameter b. This is
a consequence of the parameterization used in drc, a choice rooted in the
standard terminology used in weed science. The actual slope of the tangent
of the log-logistic model function at e is equal to:
−b
(B.3)
(d − c)/(4e)
So, apart from the sign, there is a scaling factor, depending on c, d, and e,
that converts the parameter b into the slope; therefore, the use of the phrase
“relative slope.” Note that the scaling factor may be viewed as a kind of
normalization factor involving the range of the response. As a consequence,
estimated b values often lie in the range 0.5 to 20 in absolute value, in a way
centered around 1, regardless of the assay or experiment generating the data.
The four-parameter log-logistic model may also be parameterized in other
ways than shown in Eq. (B.2). One such alternative parameterization is where
the logarithm of ED50 denoted by ẽ, say, is a model parameter instead of e
Dose-response model functions 179
as in the following equation (B.4):

d−c
f (x, (b, c, d, ẽ)) = c + (B.4)
1 + exp(b(log(x) − ẽ))
This second version of the four-parameter log-logistic model may be preferred
for dose-response analysis involving very small datasets (<15–20) where a
normally distributed parameter estimate may more reasonably be assumed
on a logarithm-transformed dose scale than the original dose scale.
The four-parameter log-logistic model is also sometimes referred to as the
(sigmoidal) E-max model (MacDougall, 2006) or Hill model, and the slope
parameter b is also sometimes referred to as the Hill coefficient or factor.
B.1.1.1 Three-parameter version

In the three-parameter log-logistic model function, the parameter c is fixed at
the value 0:
d
f (x, (b, d, e)) = (B.5)
This model function may be helpful in case there are no data to properly
estimate the lower limit. This assumption should ideally also be reasoned on
biological or toxicological grounds.
B.1.1.2 Two-parameter version

If the parameters c and d are fixed at 0 and 1, respectively, then the so-called
two-parameter log-logistic model is obtained:
1 exp(−b(log(x) − log(e)))
f (x, (b, e)) = =
1 + exp(b(log(x) − log(e))) 1 + exp(−b(log(x) − log(e)))
(B.6)
The rightmost formula shows that, for binomial data, the two-parameter log-
logistic model is equivalent to the logistic regression model with log(x) as the
only explanatory variable (in a slightly different parameterization than the
ordinary logistic regression model).
B.1.1.3 E-max and Michaelis-Menten models

Another special case is the Michaelis-Menten model. The standard two-
parameter Michaelis-Menten model is obtained by putting b = 1 and c = 0 in
Equation (B.2):
dx
f (x, (d, e)) = (B.7)
x+e
In a medical or pharmaceutical context, this model is usually called the (hy-
perbolic) E-max model (MacDougall, 2006). It is also sometimes referred to
as the rectangular hyperbola model.
The three-parameter shifted Michaelis-Menten model is defined as follows:

d−c (d − c)x
f (x, (c, d, e)) = c + =c+ (B.8)
1 + (e/x) x+e
Michaelis-Menten models are some of the best-known models of enzyme ki-
netics but they can also model numerous other biological relationships (e.g.,
Ritz and Streibig, 2008, pp. 2–3).
B.1.2 Extensions
B.1.2.1 Generalized log-logistic models
Generalized log-logistic model functions may be obtained based on cdf’s of the
Burr type III and XII distributions (Pant and Headrick, 2013). These functions
involve one additional asymmetry parameter compared to the four-parameter
log-logistic model and they include the four-parameter log-logistic model as a
special case. Inclusion of the parameter f implies that the parameter e loses
its interpretation as ED50, although ED50 may still be estimated as a derived
parameter.
The five-parameter Burr type III generalized log-logistic model function
was proposed by Finney (1979) for describing continuous dose-response data.
It is also known as the Richards model (Seber and Wild, 1989, pp. 332–333,
Sand et al., 2006). Specifically, the model function is defined as follows:
d−c
f (x, (b, c, d, e, f )) = c +
(1 + exp[b{log(x) − log(e)}])f
= c + (d − c)(1 + (x/e)b )−f (B.9)
where the parameter f , which should be positive, controls the degree of asym-
metry: f < 1 and f > 1 lead to a more rapid or a slower descent towards
the limits (for a more detailed explanation, see below under Weibull models);
see Sand et al. (2006) for an illustration of the different types of asymmetry.
For f = 1 the four-parameter log-logistic model is recovered, i.e., there is no
asymmetry. Gottschalk and Dunn (2005) provide a more detailed description
of the asymmetry.
For binomial data, the special case of the Burr type III generalized log-
logistic model obtained by fixing the lower and upper limits at 0 and 1, respec-
tively, is the so-called convenient three-parameter model proposed by Prentice
(1976); see also Shao (2000); Scholze et al. (2001).
The five-parameter Burr type XII generalized log-logistic model is defined
as follows:
f (x, (b, c, d, e, f )) = c + (d − c)(1 − (1 + (x/e)b )−f ) (B.10)
This five-parameter model does not seem to have been used in practice for
fitting dose-response data. Currently, this model is not in drc.
For binomial data there are a number of special cases: the three-parameter
model obtained by fixing the lower and upper limits at 0 and 1 (Prentice,
1976; Stukel, 1988), in another parameterization the related three-parameter
Aranda-Ordaz model (Aranda-Ordaz, 1981), and the so-called approximate
beta-Poisson model for b = 1, c = 0, and d = 1 (e.g., Namata et al., 2008).
B.1.2.2 A model with two slope parameters

The five-parameter baroflex model is a modification of the four-parameter
log-logistic model to include two slope parameters. It was originally proposed
by Ricketts and Head (1999) for studying baroreflex properties in humans. It
was proposed as an extension of the four-parameter logistic model. However,
a similar extension is possible for the four-parameter log-logistic model, mak-
ing the model more suited for describing dose-response data; this version is
implemented in drc. The model function is defined as follows:
f (x, (b1 , b2 , c, d, e))
(d − c)
=d +
1 + g(x) exp[b1 {log(x) − log(e)}] + (1 − g(x)) exp[b2 {log(x) − log(e)}]
(B.11)
where the helper function g is itself a two-parameter log-logistic function:
1
g(x) =
1 + exp[k{log(x) − log(e)}]
with slope parameter defined as k = 2b1 b2 /|b1 + b2 | (| · | denotes absolute
value).
B.1.2.3 Hormesis models

Two non-monotonous five-parameter hormesis models, which are derived from
the four-parameter log-logistic model, have been proposed by Brain and
Cousens (1989) and Cedergreen et al. (2005).
The Brain-Cousens model is defined as follows:
d − c + fx
f (x, (b, c, d, e, f )) = c + (B.12)
1 + (x/e)b
where the parameter f ≥ 0 reflects the degree of hormesis, with f = 0 im-
plying no hormetic effect as the four-parameter log-logistic model is recovered
(Equation (B.2)).
The Cedergreen-Ritz-Streibig model is defined in a similar way, but with
an exponential term instead of a linear term in the numerator:
d − c + f exp(−1/xα )
f (x, (b, c, d, e, f )) = c + (B.13)
1 + (x/e)b
where the parameter f has the same interpretation as in the Brain-Cousens
model (Cedergreen et al., 2005). The additional parameter α is usually fixed
at a value larger than 0 and less than or equal to 1. The closer α is to 1 the
steeper the ascent is towards the hormesis peak.
In contrast to the other log-logistic type models, the Brain-Cousens and
Cedergreen-Ritz-Streibig models are sensitive to the magnitudes of the doses,
which may need to be manually up- or downscaled appropriately prior to
model fitting (Belz and Piepho, 2012).
Moreover, these models can only describe decreasing trends, i.e., they are
suitable for modeling so-called inverse j-shaped hormesis models. However,
a modified Cedergreen-Ritz-Streibig model functions for u-shaped hormesis
models for describing increasing dose-response trends are also available:
d − c + f exp(−1/xα )
f (x, (b, c, d, e, f )) = d − (B.14)
1 + (x/e)b
We refer to Garnier-Laplace et al. (2010) for an application.
B.1.2.4 Two- and three-phase models

The four-parameter log-logistic model may be seen as a one-phase model re-
flecting that the response of a single biological mechanism was observed over
the dose range. In some cases, multiple mechanisms may be at work and more
complex models may be justified.
The two-phase log-logistic model is defined as follows:
d1 − c1 d2
f (x, (b1 , b2 , c1 , d1 , d2 , e1 , e2 )) = c1 + b
+ (B.15)
1 + (x/e1 ) 1 1 + (x/e2 )b2
It is a model with 7 parameters where b1 , c1 , d1 , and e1 describe the first

phase and b2 , d2 , and e2 describe the second phase, which takes over once the
first phase has reached an average response equal to d1 .
The three-phase model is defined similarly but with an additional term.
These models are extremely flexible and as a result the model fits may not al-
ways look like expected, Therefore, these models are suitable for dose-response
data where a two-phase trend is expected and where there are enough data
(e.g., Cornou et al., 2013).
B.1.2.5 Fractional polynomial models

For a binomial response, Namata et al. (2008) introduced fractional polyno-
mial log-logistic models. These models may be extended in a straightforward
way to four-parameter versions that are also suitable for describing contin-
uous dose-response data. The four-parameter model function is defined as
follows:
d−c
f (x, (b, c, d, e)) = c +
1 + exp(b(log(x + 1))p1 + e(log(x + 1))p2 )
The powers p1 and p2 need to be specified in advance (they are not estimated
from the data) and p1 has to be negative, whereas p2 has to be positive.
Following the recommendations of Royston and Altman (1994) the powers
should be chosen among the numbers –2, –1, –0.5, 0, 0.5, 1, 2, 3, e.g., some
choices are (–0.5, 0.5), (–1,1), (–1,2), (–1,3), or (–2,3).
Fractional polynomial models have been proposed as a flexible class of
candidate models for model averaging by including several choices of (p1 , p2 )
(Faes et al., 2007; Namata et al., 2008).
Finally, we note that fractional polynomial models may also be derived for
log-normal and Weibull models (Namata et al., 2008).
B.2 Log-normal models

Bruce and Versteeg (1992) introduced the four-parameter log-normal dose-
response model. We define this model using a slightly different parameteriza-
tion (following the same conventions as for log-logistic models):

f x, (b, c, d, e) = c + (d − c)Φ[b{log(x) − log(e)}] (B.16)
where Φ is the cumulative distribution function of the standard normal dis-

tribution. The parameters b, c, d, and e have the same interpretation as in
the four-parameter logistic model. Note that the parameter e has to be pos-
itive. In particular, the parameter e corresponds to ED50. Three- and two-
parameter log-normal models may be defined in the same way as log-logistic
models.
The three-parameter model was used by van der Hoeven (1997). Other-
wise, log-normal dose-responses for continuous endpoints do not appear to be
used often in practice. One reason may be that, in most statistical software
packages, they are not readily available in the general formulation given in
Equation (B.16). Moreover, as Bruce and Versteeg (1992) noted, the results
from log-normal and log-logistic models are almost identical, just like results
from logistic regression models with logit and probit link functions are very
similar (McCullagh and Nelder, 1989, p. 109).
For binomial data, the two-parameter log-normal model corresponds to the
logistic regression model with probit link function and logarithm-transformed
dose as only explanatory variable; this model is also referred to as the log-
probit model (Sand et al., 2008).
For binomial data, the probit model has been used for a long time
(Finney, 1971, Chapter 2). A recent application was reported by Wheeler
et al. (2006).
B.3 Weibull models

There are two types of Weibull models, Weibull type 1 and Weibull type 2
models. They are constructed in the same way as the Burr type III and XII
generalized log-logistic models from a suitable cdf. Sometimes the Weibull
models are confused with the Gompertz models, which constitute a different
class of models that are generally not suitable for describing dose-response
data (Ritz, 2010).
B.3.1 Weibull type 1 models

The four-parameter Weibull type 1 model is defined by the model function:
f {x, (b, c, d, e)} = c + (d − c) exp(− exp[b{log(x) − log(e)}])

= c + (d − c) exp −(x/e)b

(B.17)
where the only restriction on the parameters is that the parameter e has to
be positive.
This is the Weibull growth model considered by Piegorsch and Bailer (2005,
pp. 79–82) (in a slightly different parameterization). In contrast to the log-
logistic model, the parameter e is not equal to ED50, but it is still the location
of the inflection point of the dose-response curve. The parameter b is reflecting
the steepness of the dose-response curve, being proportional to the slope at
the dose equal to e: the larger absolute values the steeper the curve. The
asymmetry can be characterized by comparison to the symmetric log-logistic
model. The dose-response curve descends slowly from the upper limit, but the
curve approaches the lower limit rapidly. In fact, the behaviour of the Weibull
type 1 model at the upper limit is very similar to the behaviour of the log-
logistic model (which can be seen by using the approximation exp u ≈ 1 + u
for u close to 0). The figure shown by McCullagh and Nelder (1989, p. 109)
is insightful in order to understand the differences between the four classes of
models.
B.3.1.1 Exponential decay model

A special case of the Weibull type 1 model is the exponential decay model
with a non-zero lower limit:

f x, (c, d, e) = c + (d − c) exp(−x/e) (B.18)
(obtained by fixing b at 1 in the four-parameter Weibull type 1 model). Note

that by definition the parameter e is positive, ensuring that the model function
is decreasing.
In decay studies, a positive lower limit c could be justified when there

are systematic errors in the measurements of the residual concentration of a
compound in a medium, or if the degradation becomes very slow due to lack
of substrate. The exponential model with lower limit at 0 (c = 0) is used by
Stephenson et al. (2000) (in a slightly different parameterization) for analysis
of soil toxicity test data.
B.3.1.2 Other special cases

All models described by Slob (2002) are Weibull type 1 models in the sense
that they correspond to the Weibull type 1 model given by Equation (B.17) or
a special case derived from Equation (B.17) by fixing one or more parameters
at specific values. However, the parameterizations differ slightly.
A special case of the Weibull type 1 when fixing c = 0 and d = 1 could
be useful to describe quantal dose-response data, but it has not been used in
practice by ecotoxicologists. This is not entirely surprising as the correspond-
ing generalized linear model with log-log link function for binomial data is
also rare (McCullagh and Nelder, 1989, p. 108).
B.3.2 Weibull type 2 models

The second class of Weibull models is defined by the following model function:
f {x, (b, c, d, e)} = c + (d − c){1 − exp(− exp[b{log(x) − log(e)}])}
= c + (d − c){1 − exp(−(x/e)b )} (B.19)
where again the parameter e has to be positive. The parameter b has the same
interpretation as for the Weibull type 1 models.
The Weibull type 2 model differs from the Weibull type 1 model as it ex-
hibits a different form of asymmetry with a rapid change or descent from the
upper limit, but a slow approach towards the lower limit. Thus the Weibull
type 2 model is different from the log-logistic model close to the upper asymp-
tote (low doses for decreasing data), but very similar to the log-logistic model
close to the lower asymptote (McCullagh and Nelder, 1989, p. 109).
B.3.2.1 Asymptotic regression

The asymptotic regression model is a dose-response model that is obtained by
setting b = 1 in Eq. (B.19):
f (x, (c, d, e)) = c + (d − c)(1 − exp(−x/e)) (B.20)
This model may be seen as an asymmetric alternative to the Michaelis-
Menten model for dose-response data with increasing trend (Keller et al.,
2002). Asymptotic regression has also been used in animal science to describe
weight increase as a function of supplementation dose levels (Kratzer and
Littell, 2004).
B.3.2.2 Other special cases

The special case with c = 0 is sometimes called the Douglas model in pharma-
cology (Keller et al., 2002). Caux and Moore (1997) consider the case with the
constraints c = 0 and d = 100. By fixing d at 1 we obtain the first-order mul-
tistage model (Downs, 1992; Piegorsch and Bailer, 2005, p. 181); these models
are mostly suitable for binomial data. Likewise, the special case obtained for
c = 0 and d = 1 is useful for binomial data, (e.g., Meister and van den Brink,
2000; Scholze et al., 2001). Sometimes this special case is referred to as the
extreme value model (e.g., Namata et al., 2008). In another parameterization
this model is the well-known generalized linear model with the complementary
log-log link function for binomial data (McCullagh and Nelder, 1989, Chap-
ter 4). Finally, the special case obtained by fixing the three parameters c, d,
and e at the values 0, 1, and 1, respectively, is the one-hit model described by
Edler and Kopp-Schneider (1998).
B.3.2.3 Generalized Weibull-2 model

The four-parameter Weibull-2 model function may be augmented with a fifth
parameter, an additional asymmetry parameter f , in the following way:
f {x, (b, c, d, e, f )} = c + (d − c){1 − exp(− exp[b{log(x) − log(e)}])}f
= c + (d − c){1 − exp(−(x/e)b )}f (B.21)
To our knowledge this five-parameter model function has not been used much
for fitting dose-response data except for the special case with b = 1, which is
the so-called Hodgkin-Huxley model (Keller et al., 2002). This model is not
in drc.
B.4 Other types of models

B.4.1 Gamma models
The four-parameter gamma model is defined using this function:
f (x, (b, c, d, e)) = c + (d − c)Γ(bx, e, 1)
where Γ denotes the distribution function for a Γ distribution with shape
parameter e and scale parameter 1 (Wheeler and Bailer, 2009).
B.4.2 Multistage models

The four-parameter multistage model of second order is an extension of the
exponential decay model where dose squared now also enters the exponent:
f (x, c(b1 , b2 , c, d)) = c + (d − c) exp(−b1 x − b2 x2 )
The parameters b1 and b2 have to be non-negative to ensure a decreasing

dose-response model function (Piegorsch and Bailer, 2005, p. 181). In principle
higher-order terms could be included in the exponent, but they may hardly
be needed in practice. The quantal-linear and quantal-quadratic models for
binomial data are obtained by setting c = 0, d = 1, and either b2 = 0 or
b1 = 0 (Wheeler and Bailer, 2009). The exponential decay model defined in
Equation (B.18) is a special case, which is sometimes referred to as a one-stage
model (Fang et al., 2015).
B.4.3 NEC
Another built-in model is the so-called “no effect concentration” (NEC) model
proposed by Pires et al. (2002). The corresponding model function involves
four parameters:

c + (d − c) exp(b(x − e)) if x > e
f (x, (b, c, d, e)) =
d if x ≤ e
This model is a so-called threshold model such that the corresponding dose-
response curve is non-differentiable at the dose, where the constant, low-dose
level (equal to d) abruptly changes into exponential decay. This model is
similar to the hockey stick model (Environment Canada, 2005, pp. 108–110;
Ritz and Streibig, 2008, p. 41), except that linear decay has been replaced by
exponential decay.
B.4.4 Biphasic models with a peak

Recently, two biphasic dose-response models suitable for describing data ob-
tained from biosensors were suggested by Martin-Betancor et al. (2015). Like
hormesis models, these two models are useful for describing dose-response data
showing a non-monotonous behaviour: increasing towards a maximum or peak,
where the first phase changes into the second phase, and then decreasing as
doses become larger and larger.
The Gaussian model is derived from the density function of the normal
distribution:
f (x, (b, c, d, e, f )) = c + (d − c) exp(−{(x − e)/b}f /2)
This model may be used to describe biphasic dose-response data that exhibit
symmetric behaviour in terms of increase to and decrease from the maximum
response level, i.e., same behaviour for both phases.
The log-Gaussian model is defined in a similar way, but due to the log-
arithmic terms, asymmetry in the behaviour of the two phases (e.g., rapid
increase vs. slower decrease) may be incorporated:
f (x, (b, c, d, e, f )) = c + (d − c) exp(−{|{log(x) − log(e)}|/b}f /2)

For both models the parameter f may be useful for capturing varying shapes
in the dose-response data.
B.5 Fixing parameters

One key property of drc is the hierarchical structure of the built-in model
functions, which provides a convenient way to specify special cases obtained
by fixing one or more parameters a priori at certain given values. These pa-
rameters will not be estimated from the data, but will be kept fixed at the
specified values. In practice, such special cases occur quite frequently (as seen
in Table B.1 and Table B.2).
All built-in model functions in drc accept the argument fixed, which is
used for fixing parameters at given values. For instance, the general five-
parameter log-logistic model function may be used for deriving various special
cases:
• Fixing f = 1:
d−c
c+
In drc: LL.5(fixed = c(NA, NA, NA, NA, 1)) or abbreviated LL.4()
• Fixing both f = 1 and c = 0:
d
In drc: LL.4(fixed = c(NA, 0, NA, NA)) or abbreviated LL.3()

• Fixing f = 1, c = 0, and d = 1:
1
In drc: LL.4(fixed = c(NA, 0, 1, NA)) or abbreviated LL.2()
Note that NAs are used to indicate that parameters are to be estimated
from the data. Note also that model parameters come in alphabetical order:
b, c, d, e, f , which is assumed when specifying the argument fixed.
TABLE B.1
List of more commonly used model functions with the corresponding names
in drc.
Class Model name Number parms Function in drc

Log-logistic
4 LL.4()
3 LL.2()
2 LL.2()
Shifted E-max 3 MM.3()

E-max/Michaelis-
Menten 2 MM.2()
Burr type III 5 LL.5()

Log-normal
4 LN.4()
3 LN.3()
2 LN.2()
Weibull type 1
4 W1.4()
3 W1.3()
2 W1.2()
Exponential decay 3 EXD.3()

2 EXD.2()
Weibull type 2
4 W2.4()
3 W2.3()
2 W2.2()
Asymptotic regression 3 AR.3()

2 AR.2()
Gamma model 4 gammadr()
Multistage 4 multi2()
NEC
4 NEC.4()
3 NEC.3()
2 NEC.2()
Biphasic with a
peak
5 gaussian()
5 lgaussian()
TABLE B.2
List of more specialized model functions with the corresponding names in drc.
Model Specification Function in drc

Baroflex Five-parameter baro5()
Brain-Cousens
Five-parameter BC.5()
Four-parameter BC.4()
Cedergreen-Ritz-Streibig Five-parameter
α=1 CRSa.5()
α = 0.5 CRSb.5()
α = 0.25 CRSc.5()
Four-parameter
α=1 CRSa.4()
α = 0.5 CRSb.4()
α = 0.25 CRSc.4()
U-shaped Cedergreen-Ritz-Streibig Five-parameter

α=1 UCRSa.5()
α = 0.5 UCRSb.5()
α = 0.25 UCRSc.5()
Four-parameter
α=1 UCRSa.4()
α = 0.5 UCRSb.4()
α = 0.25 UCRSc.4()
Two-phase model Seven-parameter twophase()

Three-phase model Ten-parameter threephase()
Fractional polynomial Four-parameter

p1 , p2 FPL.4()
Universal response surface analysis Seven-parameter

ursa()
Appendix C
R code for plots
In this appendix we provide R code for the plotting example where more than
a few lines are needed to obtain the result. We use the following extension
packages:
library(drc)
library(devtools)
library(drcData)
install_github("DoseResponse/medrc")
library(medrc)
library(ggplot2)
C.1 Continuous dose-response data

As dose-response analysis is a type of regression analysis, it is natural to show
confidence bands around the fitted regression curve instead of providing error
bars at each dose in the dataset.
In this section we provide an example of how functionality of drc may be
used for constructing a high-quality plot, using the extension package ggplot2
(Wickham, 2016).
We need a few preparations before being able to do the actual plotting.
First we generate dose levels on the logarithmic scale as support for the
line using the function expand.grid(). Based on the model fit, the function
predict() returns predicted values with confidence intervals for all generated
dose levels and everything is gathered in the dataset newdata. Finally, plotting
doses on a logarithmic scale will result in problems unless we help by shifting
the dose 0 slightly upwards.
191
C.1.1 Ferulic acid as an herbicide

We use ggplot2 to produce a high-quality plot of the fitted dose-response curve
obtained in Subsection 1.1.3.
ryegrass.LL.4 <- drm(rootl ~ conc,

data = ryegrass,
fct = LL.4())
newdata <- expand.grid(conc = exp(seq(log(0.5), log(100),

length = 100)))
pm <- predict(ryegrass.LL.4, newdata = newdata,
interval = "confidence")
newdata$p <- pm[, 1]
newdata$pmin <- pm[, 2]
newdata$pmax <- pm[, 3]
ryegrass$conc0 <- ryegrass$conc

ryegrass$conc0[ryegrass$conc0 == 0] <- 0.5
The functionality of ggplot2 consists of different building blocks. The plot

is initialized with ggplot(). Components are added to the initial plot using
the + and the different geom functions. We use geoms to add the data points,
the predicted curve, and the ribbon presenting the confidence bands. Finally,
we specify that the x axis should be rendered on the logarithmic scale (see
Figure C.1).
C.2 Estimation of BMD

This section contains R lines for some of the plots shown in Chapter 6.
C.2.1 Pathogens in food

The R lines used for producing Figure 6.1 in Subsection 6.1.1 are shown below.
plot(pathogen.m1,
broken = TRUE, bp = 10,
xlab = "Echovirus 12 (pfu)",
ylab = "Proportion infected",
xlim = c(0, 100000),
ylim = c(0, 1),
axes = FALSE)
R code for plots 193
axis(2, c(0.0, 0.1, 0.2, 0.4, 0.6, 0.8, 1),

c("0.0", "BMR", 0.2, 0.4, 0.6, 0.8, "1.0"))
axis(1,c(90.3, 1000, 10000, 100000),

c("BMD", "1000", "10000", "100000"))
ggplot(ryegrass, aes(x = conc0, y = rootl)) +

geom_point() +
geom_line(data = newdata, aes(x = conc, y=p)) +
geom_ribbon(data = newdata,
aes(x = conc, y = p, ymin = pmin, ymax = pmax),
alpha = 0.2) +
coord_trans(x = "log") +
xlab("Ferulic acid (mM)") + ylab("Root length (cm)")
6
Root length (cm)
25 50 75100
Ferulic acid (mM)
FIGURE C.1
Four-parameter log-logistic model with confidence bands fitted to the dataset
ryegrass.
segments(1, 0.1, 90.3, 0.1)
segments(90.3, 0.1, 90.3, -100)
Likewise, the R lines for showing model averaging in Figure 6.7 in Subsec-
tion 6.3.1 look like this:
plot(pathogen.m1, broken = TRUE, bp = 10,

xlab = "Echovirus 12 (pfu)", ylab = "Proportion infected",
xlim = c(0, 100000), ylim = c(0, 1))
plot(pathogen.LN.2, broken = TRUE, bp = 10,

xlim = c(0, 100000), ylim = c(0, 1), add = TRUE,
lty = 2, type = "none")
plot(pathogen.W2.2, broken = TRUE, bp = 10,

xlim = c(0, 100000), ylim = c(0, 1), add = TRUE,
lty = 3, lwd = 2, type = "none")
legend("topleft", legend = c("LL.2", "LN.2", "W2.2"),

lty = c(1, 2, 3), lwd = c(1, 1, 2), bty = "n")
C.2.2 Toxicity of an antituberculosis drug

The R lines for producing Figure 6.8 in Subsection 6.2.2 with four fitted dose-
response curves.
plot(aconiazide.m2, broken = TRUE, type = "all",

xlab = "Aconiazide (mg/kg)", ylab = "Weight change (g)")
plot(aconiazide.LN.3, broken = TRUE, add = TRUE,

plot(aconiazide.W1.3, broken = TRUE, add = TRUE,


legend("bottomleft", legend = c("LL.4", "LN.4", "W1.4", "W2.4"),

lty = c(1, 2, 3, 4), lwd = c(1, 1, 2, 2), bty = "n")
The R lines for producing Figure 6.9 in Subsection 6.3.2 are provided below.
plot(aconiazide.m2, broken = TRUE, type = "all",

lty = 2, xlab = "Aconiazide (mg/kg)",
ylab = "Weight change (g)",
xlim = c(0, 10000), ylim = c(0, 400))
plot(aconiazide.LN.3, broken = TRUE, add = TRUE,

lty = 2, type = "none", xlim = c(0, 10000),
ylim = c(0, 400))

ylim = c(0, 400))

ylim = c(0, 400))
aconiazide.AIC < -AIC(aconiazide.m2, aconiazide.LN.3,

aconiazide.W1.3, aconiazide.W2.3)
AICWeights<-exp(-aconiazide.AIC$AIC) /
sum(exp(-aconiazide.AIC$AIC))
aconiazide.curve <- bmdMACurve(modelList = list(aconiazide.m2,

aconiazide.LN.3, aconiazide.W1.3,
aconiazide.W2.3),
modelWeights = AICWeights, 300)
g <- function(dose){aconiazide.curve$MACurve(dose) + 300}
curve(g, 1, 10000, add = TRUE, lwd = 2)
legend("bottomleft", legend = c("Individual curves",

"Model-averaged curve"),
lty = c(2, 1), lwd = c(1, 2), bty = "n")
C.3 Hierarchical nonlinear models

This section contains R lines for plots shown in Chapter 7.
C.3.1 Inhibition of photosynthesis in spinach

For the visualization of the predicted dose-response curves in the example with
spinach in Section 7.2.2, a dataset with new dose levels for both herbicides
needs to be constructed. The dplyr and tidyr packages provide an infrastruc-
ture to modify and tidy data. The pipe operator %>% applies a function on the
right-hand side to a dataframe on the left-hand side.
library(dplyr)
library(tidyr)
pdata <- spinach %>%
group_by(CURVE, HERBICIDE) %>%
expand(DOSE = exp(seq(-5, 5, length = 50)))
For each combination of the curve and herbicide indicators, a sequence of 50

new dose levels are created, which are equally spaced on the logarithmic scale.
The function predict() evaluates the predictions for the new dataframe on
the assay level and on the population level for level=0.
pdata$SLOPEind <- predict(modspinach, newdata = pdata)

pdata$SLOPE <- predict(modspinach, newdata = pdata, level = 0)
With the information about each predicted curve in the new dataframe, the
package ggplot2 is used to add the predictions to a scatterplot of the set of
observations, shown in Figure 7.2.
ggplot(spinach, aes(x = log(DOSE), y = SLOPE,

colour = HERBICIDE, group = CURVE,
shape = HERBICIDE)) +
geom_point() +
geom_line(data = pdata) +
geom_line(data = pdata, aes(y = SLOPEind), linetype = 2) +
theme_bw() +
scale_colour_grey(start = 0.1, end = 0.6) +
scale_x_continuous("DOSE",
breaks = log(c(0.01, 0.1, 1, 10, 100)),
labels = c(0.01, 0.1, 1, 10, 100))
C.3.2 Herbicides with auxin effects

For the visualization of the predicted curves of the Auxins example in Section
7.2.3, a new dataset is constructed with new dose levels at each combination
of the dummy-coded grouping, which is created in the user-defined design
matrix. The common control level is excluded from the new dataset. The
package ggplot2 is used to draw a scatterplot in Figure 7.2, adding lines for
the predictions on the individual replicate and population level.
library(dplyr)
library(tidyr)
pauxdata <- auxins %>%
group_by(replicate, herbicide, formulation,
h24D, mp, dp, comm, h24Dcomm, mpcomm, dpcomm) %>%
expand(dose=exp(seq(-4, 0, length = 25)))
pauxdata <- subset(pauxdata, formulation != "control")
pauxdata$yind <- predict(auxmod, newdata = pauxdata)

pauxdata$y <- predict(auxmod, newdata = pauxdata, level = 0)
ggplot(subset(auxins, formulation != "control"),

aes(y = dryweight, x = log(dose),
colour = herbicide, shape = herbicide)) +
geom_point() +
geom_line(data = pauxdata, aes(y = y)) +
geom_line(data = pauxdata, aes(y = yind,
group = replicate:herbicide),
lty = 2) +
facet_wrap(~ formulation) +
theme_bw() +
scale_colour_grey(start = 0.8, end = 0.2) +
scale_x_continuous("dose",
breaks = log(c(0.02, 0.05, 0.15, 0.4, 1)),
labels = c(0.02, 0.05, 0.15, 0.4, 1))
C.3.3 Drought stress resistance in Brassica oleracea

The following R code is used to construct the scatterplot in Figure 7.3 in
Section 7.2.4.
library(dplyr)
library(tidyr)
pbroc <- broccoli %>%
group_by(Stress, Genotype, ID) %>%
expand(Day = seq(1, 16, length = 25))
pbroc$ygeno <- predict(brmod, newdata = pbroc)
ggplot(broccoli,
aes(y = LeafLength, x = Day,
linetype = Stress, group = ID)) +
geom_line(colour = "grey") +
geom_line(data = pbroc, aes(y = ygeno)) +
facet_wrap(~ Genotype) +
theme_bw()
Bibliography
Aranda-Ordaz, F. J. (1981). On two families of transformation to additivity

for binary response. Biometrika 68, 357–363.
Bailer, A. J., Noble, R. B. and Wheeler, M. W. (2005). Model uncertainty and
risk estimation for experimental studies of quantal responses. Risk Analysis
25, 291–299.
Bailer, A. J. and Oris, J. T. (1997). Estimating inhibition concentrations for
different response scales using generalized linear models. Environmental
Toxicology and Chemistry 16, 1554–1559.
Beckon, W. N., Parkins, C., Maximovich, A. and Beckon, A. V. (2008). A
general approach to modeling biphasic relationships. Environmental Science
& Technology 42, 1308–1314.
Begg, C. B. and Gray, R. (1984). Calculation of polychotomous logistic regres-
sion parameters using individualized regressions. Biometrika 71, 11–18.
Beland, F. A., Dooley, K. L., Hansen, E. B. and Sheldon, W. G. (1995).
Six-month toxicity comparison of the antituberculosis drugs aconiazide and
isoniazid in Fischer 344 rats. Journal of the American College of Toxicology
14, 328–342.
Belz, R. G. and Piepho, H.-P. (2012). Modeling effective dosages in hormetic
dose-response studies. PLoS ONE 10, e33432.
Bentley, K. S., Kirkland, D., Murphy, M. and Marshall, R. (2000). Evaluation
of thresholds for benomyl- and carbendazim-induced aneuploidy in cultured
human lymphocytes using fluorescence in situ hybridization. Mutation Re-
search/Genetic Toxicology and Environmental Mutagenesis 464, 41–51.
Berkey, C. S., Hoaglin, D. C., Antczak-Bouckoms, A., Mosteller, F. and
Colditz, G. A. (1998). Meta-analysis of multiple outcomes by regression
with random effects. Statistics in Medicine 17, 2537–2550.
Brain, P. and Cousens, R. (1989). An equation to describe dose responses
where there is stimulation of growth at low dose. Weed Research 29, 93–96.
Bruce, R. D. and Versteeg, D. J. (1992). A statistical procedure for model-
ing continuous toxicity data. Environmental Toxicology and Chemistry 11,
1485–1494.
199
200 Bibliography
Buckland, S. T., Burnham, K. P. and Augustin, N. H. (1997). Model selection:

An integral part of inference. Biometrics 53, 603–618.
Buckley, B. E., Piegorsch, W. W. and West, R. W. (2009). Confidence limits on

one-stage model parameters in benchmark risk assessment. Environmental
and Ecological Statistics 16, 53–62.
Budtz-Jørgensen, E., Keiding, N. and Grandjean, P. (2001). Benchmark dose
calculation from epidemiological data. Biometrics 57, 698–706.
Cabanne, F., Gaudry, J. and Streibig, J. (1999). Influence of alkyl oleates on
efficacy of phenmedipham applied as an acetone: Water solution on Galium
aparine. Weed Research 39, 57–67.
Carroll, R. J. and Ruppert, D. (1984). Power transformations when fitting
theoretical models to data. Journal of the American Statistical Association
79, 321–328.
Carroll, R. J. and Ruppert, D. (1988). Transformations and Weighting in
Regression. Chapman & Hall, London, first edn.
Caux, P.-Y. and Moore, D. R. (1997). A spreadsheet program for estimating
low toxic effects. Environmental Toxicology and Chemistry 16, 802–806.
Cedergreen, N., Ritz, C. and Streibig, J. C. (2005). Improved empirical models
describing hormesis. Environmental Toxicology and Chemistry 24, 3166–
3172.
Chapman, G. A., Denton, D. L. and Lazorchak, J. M. (1995). Short-Term
Methods for Estimating the Chronic Toxicity of Effluents and Receiving
Waters to West Coast Marine and Estuarine Organisms. U. S. EPA.
Christensen, M. G., Teicher, H. B. and Streibig, J. C. (2003). Linking fluores-
cence induction curve and biomass in herbicide screening. Pest Management
Science 59, 1303–1310.
Cornou, C., Storm, I. M. D., Hindrichsen, I. K., Worgan, H., Bakewell, E.,
Ruiz, D. R. Y., Abecia, L., Tagliapietra, F., Cattani, M., Ritz, C. et al.
(2013). A ring test of a wireless in vitro gas production system. Animal
Production Science 53, 585–592.
Crump, K. S. (1984). A new method for determining allowable daily intakes.
Fundamental and Applied Toxicology 4, 854–871.
Davidian, M. and Giltinan, D. M. (1995). Nonlinear Models for Repeated Mea-
surement Data. Chapman & Hall, Boca Raton, Florida.
Davison, A. C. and Hinkley, D. V. (1997). Bootstrap Methods and Their Ap-

plication. Cambridge University Press, Cambridge.
Bibliography 201
Delignette-Muller, M. L., Lopes, C., Veber, P. and Charles, S. (2014a). Calcu-

lation of polychotomous logistic regression parameters using individualized
regressions. Environmental Science & Technology 48, 7544–7551.
Delignette-Muller, M. L., Lopes, C., Veber, P. and Charles, S. (2014b). Sta-
tistical handling of reproduction data for exposure-response modeling. En-
vironmetal Science & Technology 48, 7544–7551.
Demidenko, E. (2013). Mixed Models: Theory and Applications with R. Wiley

Series in Probability and Statistics. John Wiley & Sons, Inc., Hoboken, NJ,
USA, 2nd edn.
DiCiccio, T. J. and Efron, B. (1996). Bootstrap confidence intervals (with
discussion). Statistical Science 11, 189–228.
Downs, T. (1992). Biostatistical approaches for modeling U-shaped dose-
response curves and study design considerations in assessing the biological
effects of low doses. In Biological Effects of Low Level Exposures to Chem-
icals and Radiation (ed. E. J. Calabrese), 119–145. Lewis Publishers, Boca
Raton, Florida. Chapter 7.
Edler, L. and Kopp-Schneider, A. (1998). Statistical models for low dose ex-
posure. Mutation Research 405, 227–236.
Environment Canada (2005). Guidance Document on Statistical Methods for
Environmental Toxicity Tests. Environment Canada, Ottawa.
van Ewijk, P. H. and Hoekstra, J. A. (1993). Calculation of the EC50 and
its confidence interval when subtoxic stimulus is present. Ecotoxicology and
Environmental Safety 25, 25–32.
Faes, C., Aerts, M., Geys, H. and Molenberghs, G. (2007). Model averag-
ing using fractional polynomials to estimate a safe level of exposure. Risk
Analysis 27, 111–123.
Fang, C. and Zhang, D. (2014). Estimating the interaction index in drug
combination experiments. Statistics in Biopharmaceutical Research 6, 144–
153.
Fang, Q., Piegorsch, W. W. and Barnes, K. Y. (2015). Bayesian benchmark
dose analysis. Environmetrics 26, 373–382.
Faraggi, D., Izikson, P. and Reiser, B. (2003). Confidence intervals for the 50
per cent response dose. Statistics in Medicine 22, 1977–1988.
Feng, Z. and McCulloch, C. E. (1992). Statistical inference using maximum
likelihood estimation and the generalized likelihood ratio when the true pa-
rameter is on the boundary of the parameter space. Statistics & Probability
Letters 13, 325–332.
202 Bibliography
Finney, D. J. (1949). The estimation of the parameters of tolerance distribu-

tions. Biometrika 36, 239–256.
Finney, D. J. (1971). Probit Analysis. Cambridge University Press, London,

third edn.
Finney, D. J. (1979). Bioassay and the practice of statistical inference. Inter-
national Statistical Review 47, 1–12.
Fox, D. R. and Landis, W. G. (2016). Don’t be fooled - a no-observed-effect

concentration is no substitute for a poor concentration-response experiment.
Environmental Toxicology and Chemistry 35, 2141–2148.
Fox, J. F., Hogan, K. A. and Davis, A. (2017). Dose-response modeling with
summary data from developmental toxicity studies. Risk Analysis 37, 905–
917.
Garnier-Laplace, J., Della-Vedova, C., Andersson, P., Copplestone, D., Cailes,
C., Beresford, N. A., Howard, B. J., Howe, P. and Whitehouse, P. (2010).
A multi-criteria weight of evidence approach for deriving ecological bench-
marks for radioactive substances. Journal of Radiological Protection 30,
215.
Gaylor, D. W. and Slikker, W. (1990). Risk assessment for neurotoxic effects.
Neurotoxicology 11, 211–8.
Gong, X., Zeng, R., Luo, S., Yong, C. and Zheng, Q. (2004). Two new secalonic
acids from Aspergillus Japonicus and their allelopathic effects on higher
plants. In Proceedings of International Symposium on Allelopathy Research
and Application, 27-29 April, Shanshui, Guangdong, China (eds. R. Zeng
and S. Luo), 209–217.
Gottschalk, P. G. and Dunn, J. R. (2005). The five-parameter logistic: A char-
acterization and comparison with the four-parameter logistic. Analytical
Biochemistry 343, 54–65.
Hardy, A., Benford, D., Halldorsson, T., Jeger, M. J., Knutsen, K. H., More, S.,
Mortensen, A., Naegeli, H., Noteborn, H., Ockleford, C., Ricci, A., Rychen,
G., Silano, V., Solecki, R., Turck, D., Aerts, M., Bodin, L., Davis, A., Edler,
L., Gundert-Remy, U., Sand, S., Slob, W., Bottex, B., Abrahantes, J. C.,
Marques, D. C., Kass, G. and Schlatter, J. R. (2017). Update: Use of the
benchmark dose approach in risk assessment. EFSA Journal 15, 4658.
Hesterberg, T. C. (2015). What teachers should know about the bootstrap: Re-
sampling in the undergraduate statistics curriculum. The American Statis-
tician 69, 371–386.
Hoekstra, J. A. (1987). Acute bioassays with control natural mortality. Water,

Air, and Soil Pollution 35, 311–317.
Bibliography 203
van der Hoeven, N. (1997). How to measure no effect. Part III: Statistical
aspects of NOEC, ECx and NEC estimates. Environmetrics 8, 255–261.
Hothorn, T., Bretz, F. and Westfall, P. (2008). Simultaneous inference in
general parametric models. Biometrical Journal 50, 346–363.
van Houwelingen, H. C., Arends, L. R. and Stijnen, T. (2002). Advanced meth-
ods in meta-analysis: Multivariate approach and meta-regression. Statistics
in Medicine 21, 589–624.
Huber, P. J. (1981). Robust Statistics. John Wiley & Sons, Hoboken.
Inderjit, Streibig, J. C. and Olofsdotter, M. (2002). Joint action of pheno-
lic acid mixtures and its significance in allelopathy research. Physiologia
Plantarum 114, 422–428.
Izadi, H., Grundy, J. E. and Bose, R. (2012). Evaluation of the Benchmark
Dose for Point of Departure Determination for a Variety of Chemical Classes
in Applied Regulatory Settings. Risk Analysis 32, 830–835.
Jager, T., Albert, C., Preuss, T. G. and Ashauer, R. (2011). General unified
threshold model of Survival – a toxicokinetic-toxicodynamic framework for
ecotoxicology. Environmental Science & Technology 45, 2529–2540.
Jensen, S. M., Andreasen, C., Streibig, J. C., Keshtkar, E. and Ritz, C. (2017).
A note on the analysis of germination data from complex experimental
designs. Seed Science Research 27, 321–327.
Jensen, S. M. and Ritz, C. (2015). Simultaneous inference for model averaging
of derived parameters. Risk Analysis 35, 68–76.
Jensen, S. M. and Ritz, C. (2018). A comparison of approaches for simulta-
neous inference of fixed effects for multiple outcomes using linear mixed
models. Statistics in Medicine 37, 2474–2486.
Jeske, D. R., Xu, H. K., Blessinger, T., Jensen, P. and Trumble, J. (2009).
Testing for the equality of EC50 values in the presence of unequal slopes
with application to toxicity of selenium types. Journal of Agricultural,
Biological, and Environmental Statistics 14, 469–483.
Jiang, X. and Kopp-Schneider, A. (2014). Summarizing EC50 estimates from
multiple dose-response experiments: A comparison of a meta-analysis strat-
egy to a mixed-effects model approach. Biometrical Journal 56, 493–512.
Kalaian, H. A. and Raudenbush, S. W. (1996). A multivariate mixed linear
model for meta-analysis. Psychological Methods 1, 227–235.
Kang, S.-H., Kodell, R. L. and Chen, J. J. (2000). Incorporating model un-
certainties along with data uncertainties in microbial risk assessment. Reg-
ulatory Toxicology and Pharmacology 32, 68–72.
204 Bibliography
Keller, F., Giehl, M., Czock, D. and Zellner, D. (2002). PK-PD curve-fitting
problems with the Hill equation? Try one of the 1-exp functions derived
from Hodgkin, Douglas or Gompertz. International Journal of Clinical
Pharmacology and Therapeutics 40, 23–29.
Kerr, D. R. and Meador, J. P. (1996). Modeling dose response using general-
ized linear models. Environmental Toxicology and Chemistry 15, 395–401.
Keshtkar, E., Mathiassen, S. K., Beffa, R. and Kudsk, P. (2017). Seed germi-
nation and seedling emergence of blackgrass (Alopecurus myosuroides) as
affected by non–target-site herbicide resistance. Weed Science 65, 732–742.
Kociba, R., Keyes, D., Beyer, J., Carreon, R., Wade, C., Dittenber, D.,
Kalnins, R., Frauson, L., Park, C., Barnard, S., Hummel, R. and Humiston,
C. (1978). Results of a two-year chronic toxicity and oncogenicity study of
2,3,7,8-tetrachlorodibenzo-p-dioxin in rats. Toxicology and Applied Phar-
macology 46, 279–303.
Kodell, R. L. and West, R. W. (1993). Upper confidence limits on excess risk
for quantitative responses. Risk Analysis 13, 177–182.
Kooijman, S. (1981). Parametric analyses of mortality rates in bioassays. Wa-
ter Research 15, 107–119.
Kratzer, D. D. and Littell, R. C. (2004). Appropriate statistical methods to
compare dose responses of methionine sources. Poultry Science 85, 947–954.
Lindstrom, M. J. and Bates, D. M. (1990). Nonlinear mixed effects models for
repeated measures data. Biometrics 46, 673.
MacDougall, J. (2006). Analysis of dose–response Studies – Emax model. In
Dose Finding in Drug Development (ed. N. Ting). Springer.
Martin-Betancor, K., Ritz, C., Fernández-Piñas, F., Leganés, F. and Rodea-
Palomares, I. (2015). Defining an additivity framework for mixture research
in inducible whole-cell biosensors. Scientific Reports 5, 17200.
McCullagh, P. and Nelder, J. A. (1989). Generalized Linear Models. Chapman
& Hall, Boca Raton, Florida, second edn.
Meister, R. and van den Brink, P. J. (2000). The analysis of laboratory toxicity
experiments (Chapter 4). In Statistics in Ecotoxicology (ed. T. Sparks), 99–
118. John Wiley & Sons, Chichester.
Meister, R. and den Brink, P. J. V. (2000). The analysis of laboratory toxicity
experiments. In Statistics in Ecotoxicology (ed. T. Sparks), 99–118. Wiley.
Moon, H., Kim, S. B., Chen, J. J., George, N. I. and Kodell, R. L. (2013).
Model uncertainty and model averaging in the estimation of infectious doses
for microbial pathogens. Risk Analysis 33, 220–231.
Bibliography 205
Morgan, B. J. T. (1992). Analysis of Quantal Response Data. Chapman &

Hall.
Morgan, B. J. T. and Smith, D. M. (1992). A note on Wadley’s problem with
overdispersion. Applied Statistics 41, 349–354.
Murrell, J. A., Portier, C. J. and Morris, R. W. (1998). Characterizing dose-
response i: Critical assessment of the benchmark dose concept. Risk Analysis
18, 13–26.
Namata, H., Aerts, M., Faes, C. and Teunis, P. (2008). Model averaging in
microbial risk assessment using fractional polynomials. Risk Analysis 28,
891–905.
Nash, J. C. (2014). On best practice optimization methods in R. Journal of
Statistical Software 60, 1–14.
National Toxicology Program (2006). NTP technical report on the toxicology
and carcinogenesis studies of 2,3,7,8-tetrachlorodibenzo-p-dioxin (TCDD)
(CAS No. 1746-01-6) in female Harlan Sprague-Dawley rats (Gavage Stud-
ies). National Toxicology Program Technical Report Series 4–232.
Nellemann, C., Majken, D., Lam, H. R. and Vinggaard, A. M. (2003). The
combined effects of vinclozolin and procymidone do not deviate from ex-
pected additivity in vitro and in vivo. Toxicological Sciences 71, 251–262.
Normolle, D. (1993). An algorithm for robust non-linear analysis of radioim-
munoassays and other bioassays. Statistics in Medicine 12, 2025–2042.
Nottingham, Q. J. and Birch, J. B. (2000). A semiparametric approach to
analysing dose-response data. Statistics in Medicine 19, 389–404.
OECD (2006). Current Approaches in the Statistical Analysis of Ecotoxicity
Data: A Guidance to application - Annexes. Organisation for Economic
Co-operation and Development. OECD Environment Health and Safety
Publications, Series on Testing and Assessment, No. 54.
Pant, M. D. and Headrick, T. C. (2013). A method for simulating Burr Type
III and Type XII distributions through l-moments and l-correlations. ISRN
Applied Mathematics Article ID 191604.
Piegorsch, W. W. and Bailer, A. J. (2005). Analyzing Environmental Data.
John Wiley & Sons, Chichester.
Piegorsch, W. W., Xiong, H., Bhattacharya, R. N. and Lin, L. (2012). Non-
parametric estimation of benchmark doses in environmental risk assessment.
Environmetrics 23, 717–728.
Piegorsch, W. W., Xiong, H., Bhattacharya, R. N. and Lin, L. (2014). Bench-
mark dose analysis via nonparametric regression modeling. Risk Analysis
34, 135–151.
206 Bibliography
Pinheiro, J. C. and Bates, D. M. (2000). Mixed-Effects Models in S and S-

PLUS. Statistics and Computing. Springer-Verlag, New York.
Pipper, C. B., Ritz, C. and Bisgaard, H. (2012). A versatile method for confir-
matory evaluation of the effects of a covariate in multiple models. Applied
Statistics 61, 315–326.
Pires, A. M., Branco, J. A., Picado, A. and Mendonça, E. (2002). Models for
the estimation of a “no effect concentration.” Environmetrics 13, 15–27.
Prentice, R. L. (1976). A generalization of the probit and logit methods for
dose response curves. Biometrics 32, 761–768.
R Core Team (2018). R: A Language and Environment for Statistical Com-
puting. R Foundation for Statistical Computing, Vienna, Austria.
Racine, A., Grieve, A. P., Fluhler, H. and Smith, A. F. M. (1986). Bayesian
methods in practice: Experiences in the pharmaceutical industry. Applied
Racine-Poon, A. (1988). A Bayesian approach to nonlinear calibration prob-
lems. Journal of the American Statistical Association 83, 650–656.
Ricketts, J. H. and Head, G. A. (1999). A five-parameter logistic equation
for investigating asymmetry of curvature in baroreflex studies. American
Journal of Physiology (Regulatory Integrative Comp. Physiol. 46) 277, 441–
454.
Ringblom, J., Johanson, G. and Öberg, M. (2014). Current modeling practice
may lead to falsely high benchmark dose estimates. Regulatory Toxicology
and Pharmacology 69, 171–177.
Ritz, C. (2010). Towards a unified approach to dose-response modeling in
ecotoxicology. Environmental Toxicology and Chemistry 29, 220–229.
Ritz, C., Baty, F., Streibig, J. C. and Gerhard, D. (2015). Dose-response anal-
ysis using R. PLoS ONE 10, e0146021.
Ritz, C., Cedergreen, N., Jensen, J. E. and Streibig, J. C. (2006). Relative
potency in nonsimilar dose-response curves. Weed Science 54, 407–412.
Ritz, C., Gerhard, D. and Hothorn, L. A. (2013a). A unified framework for
benchmark dose estimation applied to mixed models and model averaging.
Statistics in Biopharmaceutical Research 5, 79–90.
Ritz, C., Pipper, C. B. and Streibig, J. C. (2013b). Analysis of germination
data from agricultural experiments. European Journal of Agronomy 45, 1–6.
Ritz, C. and Streibig, J. C. (2005). Bioassay analysis using R. Journal of
Statistical Software 12, 1–22.
Bibliography 207
Ritz, C. and Streibig, J. C. (2008). Nonlinear Regression with R. useR! series.

Springer, New York.
Ritz, C. and Van der Vliet, L. (2009). Handling non-normality and variance
heterogeneity for quantitative sublethal toxicity tests. Environmental Tox-
icology and Chemistry 28, 2009–2017.
Royston, P. and Altman, D. G. (1994). Regression using fractional polynomi-
als of continuous covariates: Parsimonious parametric modelling. Applied
Sand, S., Filipsson, A. F. and Victorin, K. (2002). Evaluation of the benchmark
dose method for dichotomous data: Model dependence and model selection.
Regulatory Toxicology and Pharmacology 36, 184–197.
Sand, S., von Rosen, D., Victorin, K. and Filipsson, A. F. (2006). Identification
of a critical dose level for risk assessment: Developments in benchmark dose
analysis of continuous endpoints. Toxicological Sciences 90, 241–251.
Sand, S., Victorin, K. and Filipsson, A. F. (2008). The current state of knowl-
edge on the use of the benchmark dose concept in risk assessment. Journal
of Applied Toxicology 28, 405–421.
Scholze, M., Boedeker, W., Faust, M., Backhaus, T., Altenburger, R. and
Grimme, L. H. (2001). A general best-fit method for concentration-response
curves and the estimation of low-effect concentrations. Environmental Tox-
icology and Chemistry 20, 448–457.
Seber, G. A. F. and Wild, C. J. (1989). Nonlinear Regression. John Wiley &
Sons, New York.
Shao, K. (2012). A comparison of three methods for integrating historical
information for Bayesian model averaged benchmark dose estimation. En-
vironmental Toxicology and Pharmacology 34, 288–296.
Shao, Q. (2000). Estimation for hazardous concentrations based on NOEC
toxicity data: An alternative approach. Environmetrics 11, 583–595.
Slob, W. (2002). Dose-response modeling of continuous endpoints. Toxicolog-
ical Sciences 66, 298–312.
Stephenson, G. L., Koper, N., Atkinson, G. F., Solomon, K. R. and Scrog-
gins, R. P. (2000). Use of nonlinear regression techniques for describing
concentration-response relationships of plant species exposed to contami-
nated site soils. Environmental Toxicology and Chemistry 19, 2968–2981.
Streibig, J. C. (1981). A method for determining the biological effect of her-
bicide mixtures. Weed Science 29, 469–473.
208 Bibliography
Streibig, J. C. (1983). Joint action of root-absorbed mixtures of dpx-4189 and

linuron in Sinapis alba l. and barley. Weed Research 23, 3–9.
Streibig, J. C. (1987). Joint action of root-absorbed mixtures of auxin herbi-
cides in Sinapis alba L. and barley (Hordeum vulgare L.). Weed Research
27, 337–347.
Streibig, J. C., Dayan, F. E., Rimando, A. M. and Duke, S. O. (1999). Joint
action of natural and synthetic photosystem II inhibitors. Pesticide Science
55, 137–146.
Streibig, J. C. and Jensen, J. E. (2000). Actions of herbicides in mixtures.
In Herbicides and Their Mechanisms of Action. Sheffield Academic Press,
Sheffield.
Streibig, J. C., Rudemo, M. and Jensen, J. E. (1993). Dose-response curves
and statistical models. In Herbicide Bioassays (eds. J. C. Streibig and
P. Kudsk), 29–55. CRC Press, Boca Raton.
Strodl Andersen, J., Holst, H., Spliid, H., Andersen, H., Baun, A. and Nyholm,
N. (1998). Continuous ecotoxicological data evaluated relative to a control
response. Journal of Agricultural, Biological, and Environmental Statistics
3, 405–420.
Stromberg, A. J. (1993). Computation of high breakdown nonlinear regression
parameters. Journal of the American Statistical Association 88, 237–244.
Stukel, T. A. (1988). Generalized logistic models. Journal of the American
Statistical Association 83, 426–431.
Szöcs, E. and Schäfer, R. B. (2015). Ecotoxicology is not normal. Environ-
mental Science and Pollution Research 22, 13990–13999.
Trajstman, A. C. (1989). Indices for comparing decontaminants when data
come from dose-response survival and contamination experiments. Applied
U. S. EPA (2012). Benchmark Dose Technical Guidance. 1–99.
Uptmoor, R., Osei-Kwarteng, M., Guertler, S. and Stuetzel, H. (2009). Mod-
eling the effects of drought stress on leaf development in a Brassica oleracea
doubled haploid population using two-phase linear functions. Journal of the
American Society for Horticultural Sciences 134, 543–552.
van der Vaart, A. W. (1998). Asymptotic Statistics. Cambridge University
Press, Cambridge.
Van der Vliet, L. and Ritz, C. (2013). Statistics for analyzing ecotoxicity
test data. In Encyclopedia of Aquatic Ecotoxicology (eds. J. Férard and
C. Blaise), 1081–1096. Springer, New York.
Bibliography 209
Venables, W. N. and Ripley, B. D. (2002). Modern Applied Statistics with S.

Springer, New York, fourth edn. ISBN 0-387-95457-0.
Viechtbauer, W. (2010). Conducting Meta-Analyses in R with the metafor

Package. Journal of Statistical Software 36, 1–48.
Vindimian, E., Robaut, C. and Fillion, G. (1983). A method for cooperative
or noncooperative binding studies using nonlinear regression analysis on a
microcomputer. Journal of Applied Biochemistry 5, 261–268.
Vølund, A. (1978). Application of the four-parameter logistic model to bioas-
say: Comparison with slope ratio and parallel line models. Biometrics 34,
357–365.
Weisberg, S. (2005). Applied Linear Regression. John Wiley & Sons, New
York, third edn.
West, R. W., Piegorsch, W. W., Peña, E. A., An, L., Wu, W., Wickens, A. A.,
Xiong, H. and Chen, W. (2012). The impact of model uncertainty on bench-
mark dose estimation. Environmetrics 23, 706–716.
Wheeler, M. W. and Bailer, A. J. (2007). Properties of model-averaged
BMDLs: A study of model averaging in dichotomous response risk esti-
mation. Risk Analysis 27, 659–670.
Wheeler, M. W. and Bailer, A. J. (2009). Comparing model averaging with
other model selection strategies for benchmark dose estimation. Environ-
mental and Ecological Statistics 16, 37–51.
Wheeler, M. W., Bailer, A. J. et al. (2008). Model averaging software for
dichotomous dose response risk estimation. Journal of Statistical Software
26, 1–15.
Wheeler, M. W., Park, R. M. and Bailer, A. J. (2006). Comparing median
lethal concentration values using confidence interval overlap or ratio tests.
Environmental Toxicology and Chemistry 25, 1441–1444.
White, H. (1981). Consequences and detection of misspecified nonlinear regres-
sion models. Journal of the American Statistical Association 76, 419–433.
White, H. (1996). Estimation, Inference and Specification Analysis. Cam-
bridge University Press, Cambridge.
WHO (2005). The WHO Recommended Classification of Pesticides by Hazard
and Guidelines to Classification 2004. World Health Organisation, Geneva.
Wickham, H. (2016). ggplot2: Elegant Graphics for Data Analysis. Springer-
Verlag New York. URL http://ggplot2.org.
210 Bibliography
Yu, Z.-F. and Catalano, P. J. (2005). Quantitative risk assessment for mul-
tivariate continuous outcomes with application to neurotoxicology: The bi-
variate case. Biometrics 61, 757–766.
Zeileis, A. (2006). Object-oriented computation of sandwich estimators. Jour-
nal of Statistical Software 16, 1–16.
Zeileis, A. and Hothorn, T. (2002). Diagnostic checking in regression relation-
ships. R News 2, 7–10.
Zhu, Y., Wang, T. and Jelsovsky, J. Z. (2007). Bootstrap estimation of bench-
mark doses and confidence limits with clustered quantal data. Risk Analysis
27, 447–465.
Index
Brassica oleracea, 157 Binary data, 164

Galium aparine, 31 Binomial data, 164
Lemna minor, 64 Binomial response, 120
Oncorhynchus mykiss, 6 Biphasic model, 187, 189
Sinapis alba, 35 Blackgrass, 109
Echovirus 12, 120, 136 Bonferroni adjustment, 61
Bootstrap, 28, 121, 140
Abbott’s formula, 49 Bovine tuberculosis, 79
Acetolactate synthase inhibitors, 20 Box-Cox transformation, 17, 166
Aconiazide, 133, 140 Brain-Cousens model, 181, 190
Acute inhalation, 44 Four parameters, 23, 71
Added risk, 120 Burr type III generalized log-logistic
Additional risk, 120 model, 180, 189
Adverse event, 128 Burr type XII generalized log-logistic
AIC, 163 model, 184
AIC weights, 136
Approximate F -test, 34 Cadmium chloride, 104
Approximate beta-Poisson model, Calibration experiment, 26
181 Carbendazim, 124
Aranda-Ordaz model, 181 Cedergreen-Ritz-Streibig model, 181,
Arbovirus, 90 190
Asymptotic regression model, 185, Chickweed, 97
189 Chloroacetamide, 49
Auxins, 154 Chlordane, 73
Chromosomal damage, 124
Back-transformation, 109 Confidence band, 102, 119, 172
Backfitting, 27 Confidence interval, 101, 102, 119,
Background level, 119 127, 172
Background risk, 120 Adjusted, 60
Baroflex model, 181, 190 Constrained estimation, 168
BCa, 123, 141 Contrast matrix, 116
Benchmark dose, 119 Copper, 129
Benchmark dose lower limit, 119 Count data, 63, 164
Benchmark response, 119
Benchmark risk, 119 Daphnia, 104
Bias-variance trade-off, 52, 63 Decontaminants, 79
BIC, 163 Delta method, 107, 173
211
212 Index
Derived parameter, 48, 173, 174 Inverse regression, 134, 138, 172
Diquat, 20
Douglas model, 186 J-shaped model, 182
Joint model, 36, 79
E-max model, 179, 189
Earthworms, 49, 52 Least squares estimation, 161
Ecotoxicology, 6 Lettuce, 23
Effective dose, 3, 6, 119, 128, 172, Likelihood ratio test, 59, 163
173, 178 Link function, 70
Absolute, 51 Log-Gaussian model, 187
Interpretation, 68 Log-logistic model, 96, 178, 189
Estimation, 161 Four parameters, 3, 9, 13, 20,
Excess risk, 120 21, 32, 36, 124, 130, 133
Exponential decay model, 7, 184, 189 Three parameters, 27, 52, 65,
Extra risk, 120 74, 98, 112, 135, 140, 147,
Extrapolation, 121 151, 155, 158
Two parameters, 44, 47, 58, 86,
Fieller interval, 175 91, 105, 121, 126, 136
Fixing parameters, 188 Log-normal model, 183, 189
Fluoranthene, 54 Three parameters, 50, 140
Fractional polynomial, 182, 190 Two parameters, 54, 136
Fronds, 64 Logarithm transformation, 15, 107
Fungicide, 124 Logarithmic axis, 4
Logistic model, 178
Gamma model, 186, 189 Logistic regression model, 46, 179
Generalized linear models, 43, 63, 69 Lower limit, 6, 19
Generalized log-logistic model, 180 Lymphocytes, 124
Generalized nonlinear model, 50, 52
Genetic study, 124 Maximum likelihood estimation, 163
Germination, 95, 97, 103, 109 Median event time, 96
Giant Kelp, 129 Median germination time, 96
Global test, 34 Meta-analytic approach, 39, 105,
Gompertz model, 184 114, 127
Michaelis-Menten model, 179, 189
Hierarchical design, 109, 113, 145 Model averaging, 135, 174, 183
Hill model, 179 Model checking, 8, 32
Hockey stick model, 187 Model misspecification, 11, 77
Hodgkin-Huxley model, 186 Model reduction, 59
Hormesis, 23 Model selection, 112, 135
Hormesis model, 173, 181, 190 Model-averaged curve, 142
Hormetic effect, 71 Monitoring intervals, 95
Hybrid approach, 129 Multinomial data, 85
Multistage model, 186, 189
In vitro study, 124
Insecticide residues, 86 Naive standard errors, 13
Interpolation, 135 Natural mortality, 49
Index 213
NEC model, 187, 189 Safe dose, 121

Negative binomial distribution, 165 Sandwich variance estimator, 12,
Negative binomial model, 77, 82 100, 101, 167
Nonlinear calibration, 26 Selectivity index, 174
Nonlinear least squares, 2, 162 Selenium, 57
Self-starter function, 171
Observation periods, 75 Separate models, 36
Odds, 43 Starting values, 171
Offspring, 69, 73 Sub-experiments, 31, 35
One-hit model, 186 Survival analysis, 95
One-stage model, 187
Over-dispersion, 43, 64, 77, 80, 164 TCDD, 126
Three-phase model, 182, 190
Pairwise comparisons, 56, 116, 174 Threshold model, 187
Parameterization, 46 Time to death, 104
Pathogen, 120, 136 Time-to-event-response data, 95,
Percentile, 123, 129, 141, 172 109, 165
Phenmedipham, 31 Toxicity test, 44
Poisson distribution, 164 Transform-both-sides approach, 15
Pre-processing, 49 Transformation, 109, 119, 162,
Prediction, 102 166
Prediction interval, 172 Trichotomous data, 85
Proportional hazards, 89 Tumor incidence, 126
Proportional odds, 89 Tumor incidence – estimating EC
values, 47
Quadratic polynomial, 69 Two-phase model, 182, 190
Quantal-linear model, 187 Two-stage estimation, 168
Quantal-quadratic model, 187 Two-step approach, 103, 104
Random effects, 114, 115, 146, 150, U-shaped model, 182
152 Unexposed population, 119, 129
Red fescue, 20
Relative potency, 37, 89, 91, 174 Variance heterogeneity, 162, 166,
Replicated experiments, 54 167
Replication, 2 Variance-covariance, 146, 147, 152,
Residual plot, 98, 132, 166 163, 167
Response surface analysis, 190
Richards model, 180 Wadley’s problem, 79, 80
Right censoring, 95, 165 Weibull model, 184, 189
Risk assessment, 119 Three parameters, 80, 140
Robust estimation, 166 Two parameters, 136
Robust standard errors, 12, 19, 43, Weight change, 133
57, 78, 81, 90, 133 Weighted estimation, 127
Ryegrass, 20 Weighted regression, 127, 162

(Chapman & Hall - CRC The R Series) Christian Ritz - Signe Marie Jensen - Daniel Gerhard - Jens Carl Streibig - Dose-Response Analysis Using R-CRC Press (2020)

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

(Chapman & Hall - CRC The R Series) Christian Ritz - Signe Marie Jensen - Daniel Gerhard - Jens Carl Streibig - Dose-Response Analysis Using R-CRC Press (2020)

Uploaded by

Copyright:

Available Formats

Dose-Response

No claim to original U.S. Government works

Printed on acid-free paper

International Standard Book Number-13: 978-1-138-03431-0 (Hardback)

Trademark Notice: Product or corporate names may be trademarks or registered trade-

Library of Congress Cataloging-in-Publication Data

Names: Ritz, Christian, author.

Visit the Taylor & Francis Web site at

and the CRC Press Web site at

2 Binary and binomial dose-response data 43

3 Count dose-response data 63

3.1.3 More counting offspring: Varying observation periods 73

4 Multinomial dose-response data 85

6 Benchmark dose estimation 119

7 Hierarchical nonlinear models 145

Appendix A Estimation 161

A.2.2.1 The Poisson distribution . . . . . . . . . . . 164

Appendix B Dose-response model functions 177

Appendix C R code for plots 191

(non)linear regression, and parametric survival analysis. Perhaps this is one

What this book is and is not about

How to read the book

Above we praised the fact that modern dose-response analysis may be

In this chapter, we show examples of how to analyze dose-response data that

1.1 Analysis of single dose-response curves

1.1.1 Inhibitory effect of secalonic acid

Dose-response data were obtained from an experiment assessing the inhibitory

1.1.1.1 Fitting the model

secalonic.LL.4 <- drm(rootl ~ dose,

Note that upon execution of the above R lines, no output is produced.

model reduction is not meaningful and should be avoided. Assuming a lower

1.1.1.2 Estimation of arbitrary ED values

ED(secalonic.LL.4, c(10, 20))

The output shows estimated ED values and corresponding (approximate) stan-

1.1.2 Data from a fish test in ecotoxicology

toxicity experiment with 7 concentrations: 0 and 6 non-zero concentrations.

2.0 2.2 2.4 2.6 2.8

1.1.3 Ferulic acid as an herbicide

1.0 Normal Q−Q Plot

ryegrass.LL.4 <- drm(rootl ~ conc,

observation further by means of graphical model checking: A residual plot

## d:(Intercept) 7.79296 0.15311 50.8976 < 2.2e-16 ***

## Estimate Std. Error t-value p-value

1.1.4 Glyphosate in barley

barley.LL.4 <- drm(weight ~ Dose,

barley.LL.4.bc <- boxcox(barley.LL.4,

## 0.1121892 (14 degrees of freedom)

In summary both the transform-both-sides approach and the use of robust

1.1.5 Lower limits for dose-response data

## dose biomass day

We fit a four-parameter log-logistic model to the subset of data containing

ryegrass2.LL.4 <- drm(biomass ~ dose,

## dose day biomass

abline(h = mean(subset(ryegrass2, day == "0")[["biomass"]]),

We fit a four-parameter log-logistic model.

red.fescue.LL.4 <- drm(biomass ~ dose,

abline(h = mean(subset(red.fescue, day =="0")[["biomass"]]),

1.1.6 A hormesis effect on lettuce growth

## Warning in xy.coords(x, y, xlabel, ylabel, log): 2 x values

0.5 2.0 5.0 20.0 100.0

Isobutylalcohol concentration (mg/L)

log-logistic models: b is proportional to the slope at the concentration equal

lettuce.BC.4 <- drm(weight~conc,